Every year, cybersecurity vendors, analysts, and research firms publish a flood of reports examining how attackers operate, where defenses fall short, and what breaches really cost. Read one at a time, each report offers a useful snapshot. Read together, they paint a far more sobering picture of the threat landscape. In 2026, the numbers continue to point in one direction: attackers are moving faster, leveraging AI to scale their campaigns, and exploiting identity gaps, cloud misconfigurations, and third-party relationships that many organizations still struggle to see, let alone secure.
This roundup brings together some of the most alarming statistics from research published throughout 2026 by a wide range of companies across the industry. We have pulled out the figures that stand out, whether they involve ransomware, phishing, software supply chain risk, or the widening gap between the pace of threats and the pace of response. Consider this your one-stop reference for the data behind the headlines, whether you are building a business case for security investment, preparing a board presentation, or simply trying to understand where the risks are heading next.
++
Object First released its second annual Under Pressure: Mental Health Challenges in IT & Security Roles survey. Key findings include:
- Leading causes of stress: 50% cited the risk of cyberattacks, and 50% pointed to high workloads or understaffed teams. Pressure to maintain uptime followed at 44%, gaps in backup and recovery effectiveness at 41%, and lack of proper tools or resources at 33%.
- AI is exacerbating this strain: 91% of IT and cybersecurity professionals are uncomfortably stressed at work (up from 84% in 2025). 70% report that the rise of AI-powered cyber threats has affected their workplace stress as less than 1 in 4 (24%) believe their organization is ready to handle such threats.
- IT teams feel personally responsible for organizational resilience: 42% feel pressure from leadership to “fix everything” in the aftermath of a security incident, and 19% feel “hopeless and overwhelmed” during and after an incident.
- Organizational support is weakening as turnover risk rises: 59% feel their companies don’t consistently prioritize employee wellbeing and mental health, and 78% said workplace stress negatively affects their job performance. At the same time, 60% have considered or are actively looking for new jobs due to work-related stress.
And according to the company’s September research with Omdia, more organizations are falling victim to successful ransomware attacks, and recovery outcomes are worsening compared to last year. Eighty-three percent of organizations have fallen victim to a successful ransomware attack in the last 24 months, and of those, only 39% were able to recover at least 75% of their data.
++
Jim McGann, CMO of Index Engines
One of the biggest concerns with ransomware today is how little time organizations may have to respond before the damage is already done. The speed of an attack can quickly outpace the steps needed to investigate an alert, confirm a threat and take action. And increasingly, the damage itself may not be immediately obvious.
Research from our CyberSense Research lab examined 1,064 unique ransomware strains in the first half of 2026 highlights several concerning findings:
- 97,321 files per hour: The median rate of file corruption observed in controlled testing, with 10,000 files potentially affected in just six minutes.
- 16,980 files: The median number of files damaged per incident during testing.
- 39%: The percentage of strains exhibiting directory-entry corruption, compared with just 18.3% using full encryption.
- 31.4%: The percentage using partial encryption, which can make files unusable without encrypting their entire contents.
What makes these findings particularly concerning is that organizations are dealing with two challenges at once: less time to stop an attack and fewer obvious signs that data has been compromised. Preventing an attack remains important, but knowing whether data can actually be trusted afterward is just as critical to recovery.
++
What does ExaGrid find scary this time of the year?
- I can hear the threat actors howling at night as most IT organizations do not use 2FA for all their IT system passwords!
- What is scary or spooky? Having no idea if the data you have backed up is recoverable!
- Network-facing backup storage is like leaving the door open in a haunted house!
- Sleep with one eye open if an old vampire lives nearby, and also if your backup storage is so old that the drives are ready to fail!
++
New research from ControlUp, based on an analysis of more than 1 million real-world enterprise help desk tickets, reveals some frightening realities about IT support. Key findings include:
- Sign-in nightmares: Password resets, account lockouts and MFA issues accounted for up to 53% of help desk tickets at individual enterprises.
- Network outages that won’t die: At one retailer, whole-site network outages took a median of almost a week to resolve.
- Virtual desktop horror: Virtual desktop problems appeared in 55% of organizations analyzed. At one bank, 51% of virtual desktop incidents left employees unable to get a working desktop.
- The MFA curse: MFA enrollment problems appeared in 40% of organizations, often triggered by employees replacing their phones.
- Machines creating more work: Machine-generated tickets appeared in half of organizations studied, accounting for 25% to 50% of all tickets in some environments.
- Help desk overload: At one organization, phone calls accounted for a staggering 99% of recorded support contacts.
The scariest finding: The tickets that are easiest to close aren’t necessarily the ones causing the most damage. While password resets can take minutes, infrastructure failures can disrupt operations for days.
++
Keeper Security’s Insight Report Identity Security at Machine Speed:
- Keeper Security’s research from 2026 reveals more than 72% of organizations take hours, days or even weeks to detect credential misuse. For security teams, the scariest horror story this Halloween isn’t an exotic new exploit – it’s the state of threat detection in the modern enterprise. The reality is that the most dangerous attacks today are a result of malicious actors logging in rather than breaking in. When adversaries use valid credentials, they can blend directly into normal administrative traffic. That lack of visibility is why detection times drag on, giving attackers ample time to execute lateral movement, exfiltrate data and modify critical configurations.
- Security operations cannot rely solely on post-event log reviews to catch credential abuse. Too much of the industry is stuck in a forensic mindset, chasing alerts long after the initial intrusion. By the time an anomaly surfaces in a traditional SIEM, the blast radius is already massive. Defense requires shifting from historical auditing to real-time intervention by removing standing administrative rights, enforcing just-in-time access and continuously monitoring privileged sessions as they happen. Shrinking the gap between initial credential misuse and active session termination requires operating at machine speed to limit an intrusion before it turns into a catastrophic breach.
++
New Sumo Logic Security Operations Report Finds Two-Thirds of Security Leaders Lack Integrated Security Tooling
The Sumo Logic 2026 Security Operations Insights report surveyed more than 500 IT and security leaders. Key findings include:
- 90% of security operations leaders say supporting data sources from multi-cloud and hybrid-cloud environments is very or extremely important for their SIEM, highlighting the continued need for data pipeline management.
- Only 51% say their current SIEM is very effective at reducing mean time to detect and respond to threats. And just 52% are very confident their current SIEM can scale to meet future security and cloud operations needs.
- 90% of security leaders say AI/ML is extremely or very valuable in reducing alert fatigue and improving detection accuracy. Yet their most common AI use cases focus on basic tasks like threat detection. These findings indicate that AI adoption isn’t as widespread through advanced security workflows as marketing narratives often suggest.
- 93% of enterprise organizations use at least three security operations tools, and 45% use six or more. It’s no surprise that over half (55%) of respondents report having too many point solutions in their security stack.
- 80% of enterprise organizations say security and DevOps use shared observability tools, but only 45% say the two teams are very aligned on tooling and workflows. 100% say a unified platform for logs, metrics, and traces would be valuable for their security and DevOps teams.
- 70% of respondents say they’ve fully or mostly automated their threat detection and response process, with 25% reporting it’s fully automated. Those who rely on a mostly or fully manual process are in the extreme minority.
++
Gigamon released its 2026 Hybrid Cloud Security Survey, along with deeper reports into CISO insights and financial services trends, revealing how quickly AI is reshaping the cybersecurity threat landscape and exposing gaps in visibility and governance. You can’t secure what you can’t see, and as AI adoption accelerates, that blind spot is only getting bigger.
According to its findings:
- 65% of organizations experienced a breach in the past year, with 83% of those reporting AI involvement.
- 80% of CISOs cite inadequate governance around unsanctioned AI use as a major data security challenge.
- 98% of breached financial services organizations reported material business impact, including financial losses, data loss or regulatory consequences.
++
Crystal Morin, Senior Cybersecurity Strategist at Sysdig, published Four ways AI has Fundamentally Changed the Threat Landscape in 2026, a synthesis of research from Sysdig’s Threat Research Team (TRT) on how agentic AI is reshaping attacks. TRT investigates developments in the threat landscape, publishing detection and mitigation guidance. It documented JADEPUFFER, the first known case of agentic ransomware, and coined the term LLMjacking, which describes attackers using stolen cloud credentials to access AI models. The research found that AI agents are now running attacks end to end, and that attackers go after new flaws within hours of disclosure.
Key findings include:
- In JADEPUFFER, the first documented agentic ransomware, the agent fixed a failed login in 31 seconds with 15 lines of code, then encrypted over 1,300 configuration entries.
- TRT caught an attacker using someone else’s unprotected AI server to power a hacking tool that ran its password-and-key extraction step well over 100 times. It’s the newest form of LLMjacking: attackers once stole AI access to run up the victim’s bill, and now they use it as the free brain of the attack
- Attackers began going after newly disclosed flaws within hours, sometimes before a CVE was assigned:
- Under 4 hours: PraisonAI, a tool for building teams of AI agents. The flaw let anyone use it without a password.
- 12 hours: LMDeploy, software that runs AI models. The flaw let attackers trick the server into making requests for them.
- 20 hours: Langflow, a tool for building AI agents and apps. The flaw let attackers run their own commands on it.
- 36 hours: LiteLLM, a gateway that connects apps to AI models. The flaw let attackers slip malicious database commands through its login path.
- In about an hour, an AI agent got into a company’s systems through a flaw in marimo, a tool programmers use to write and run code in a notebook, and copied an internal database in four steps, with no human steering it.
++
Fenix24 released its first-ever State of Recoverability Report, built from the data of more than 500 real ransomware recoveries and 800 client engagements. The pattern they uncovered is that recovery plans fail in the same way, often looking solid on paper but fall apart the moment a real attacker is in the environment.
The gap between the recovery plan and what recovery actually requires is where companies lose weeks, not hours, and Fenix24’s data shows exactly where that happens:
- Only 4 of 800+ engagements came close to companies’ own 24–48 hour recovery targets, with none hitting full operational capacity for weeks.
- 99.2% of clients have no documented plan to recover identity systems, and the few plans that existed didn’t survive contact with the attacker.
- 95% have no meaningful MFA on infrastructure consoles, vs. just 15% lacking it at the network perimeter, highlighting that the front door is guarded, the house isn’t.
- Zero clients have ever arrived with a complete map of their own application dependencies.
- 38% of “surviving” backups still couldn’t carry a recovery.
++
Here’s three sets of scary stats to consider:
1. CYGNVS Cyber Resilience Benchmark: More than 6 in 10 organizations say incident volume has risen over the past 12 months, and 91% said their board became more involved. Yet 84% review their incident response plans once a year or less and 17% do not run tabletop exercises at all, a rhythm set when critical incidents were rare.
2. Fleet Device Management (Road to AI in IT): 79% of organizations take more than a day to deploy critical security patches, even as attackers increasingly exploit vulnerabilities within hours. Six in 10 organizations still lack complete visibility across their device fleets.
3. HYPR State of HR Identity Fraud Detection: 98% of HR teams have experienced candidate fraud firsthand. 68% of hiring fraud is detected by human instinct – not process nor tooling. 24% take 1–3 months to fully resolve a single fake hire.
++
Check Point Research recently released its 2026 AI Security Report and 2026 Cloud Security Report, uncovering some alarming findings about how AI is accelerating cyber threats while organizations struggle to keep their security defenses up to speed.
A couple of the scariest findings include:
- 88,000 lines of malware framework code developed by one person in under a week: Check Point Research uncovered VoidLink, a sophisticated Linux malware framework developed by a single person using AI in less than a week, demonstrating how quickly AI can accelerate the development of complex cyber threats.
- 70% of organizations are running GenAI in production, but only 26% have the security architecture to enforce their own strategies: Check Point’s 2026 Cloud Security Report reveals a significant gap between AI adoption and security readiness.
++
GitLab’s 2026 AI Accountability Report, conducted by The Harris Poll, surveying 1,528 DevSecOps professionals across six countries found that while AI coding tools are accelerating software development, organizations are struggling to keep pace with the security, governance, and traceability challenges introduced by AI-generated code.
Key findings include:
- 80% of DevSecOps professionals say their organization adopted AI coding tools faster than it built policies to govern them
- 92% report facing at least one governance challenge with AI-generated code, yet 78% say their organization believes its governance is adequate despite incomplete controls
- 87% are confident their team could determine within 24 hours whether AI-generated code contributed to a production incident. Yet 34% of organizations that had an incident in the past year could not make that determination.
- 43% cite difficulty distinguishing AI-generated code from human-written code as a barrier to full code traceability, while 40% point to fragmented toolchains that don’t share context
- 85% agree that organizations unable to trace or govern AI-generated code will face growing operational and security risks
For many teams, the most unsettling code in production has no answer to three questions. Where did it come from, what was it meant to do, and who owns it now?
++
Talkdesk’s “The State of Agentic Automation in CX” report found that companies are deploying AI in customer experience faster than they can make it work:
- While 98% of organizations have deployed AI in their customer journey, only 15% combine agentic AI with cross-departmental orchestration to resolve customer needs end-to-end.
- 85% of organizations lack the orchestration needed to connect AI agents, human teams, data, and workflows across enterprise systems to turn a customer’s request into a fully executed resolution, and only 5% can quantify AI’s impact on business outcomes.
- Nearly one in five organizations already view AI agents more as labor than technology, and 99% believe a hybrid workforce of AI and people delivers beneficial value.
++
TELUS Digital’s GenAI Safety Model Benchmark tested 34 models (like Claude, GPT, and Gemini) from 10 global providers across more than 620,000 attack simulations. It found:
- Two structural factors were the strongest predictors of AI safety: how a model reasons and how large it is. “Reasoning” models were vulnerable to just 19.9% of attacks compared to 55.1% for non-reasoning models, and smaller models were consistently the most exploitable.
- Some models engage with harmful requests more than 90% of the time.
++
Gravitee’s 2026 State of AI Agent Security Report highlights some concerning gaps in enterprise AI agent security as organizations rapidly expand their deployments:
- 54% of organizations experienced or suspected an AI agent security or data privacy incident in the past 12 months, including nearly 35% that confirmed an incident.
- 48% of AI agents in production are operating without active security monitoring or governance.
- 85% of organizations lack clearly defined accountability for AI agent behavior, with just 7.2% assigning formal responsibility to a named individual.
- 80% of organizations deploy at least some AI agents before they are fully secured and governed.
- 81% of technology leaders feel pressure to deploy AI agents quickly, even when security or governance measures are incomplete.
++
2026 threat data from Barracuda Research reveals how cybercriminals are adapting phishing attacks to deceive both people and the AI-powered tools designed to protect them. From hiding malicious content in plain sight to deploying industrial-scale phishing operations, the findings show how attackers are exploiting trusted communication channels and increasingly sophisticated security defenses.
Key findings include:
- More than 1 million phishing attacks using a technique called text salting were detected by Barracuda researchers beginning in April 2026. Attackers concealed harmless-looking text within malicious emails to confuse traditional and AI-powered security tools, potentially causing phishing attempts to appear legitimate.
- 90% of high-volume phishing campaigns in 2025 used phishing-as-a-service kits, giving cybercriminals access to sophisticated attack tools and automation without requiring advanced technical expertise.
- 70% of malicious PDFs analyzed in Barracuda’s 2026 Email Threats Report contained QR codes directing victims to phishing websites, turning a familiar document format into a vehicle for credential theft.
- 34% of companies reported experiencing at least one account takeover incident every month, highlighting how frequently attackers gain access to legitimate business identities and accounts.
++
A single stolen password could put multiple banks at risk. Flare Research’s October 2026 report, Cost of Convenience: Stolen Financial Services Credentials Up by 56% Across North America, reveals a sharp rise in compromised banking and financial accounts, with hundreds of thousands of stolen credentials circulating on underground forums. More concerning, researchers found that some infected devices held credentials for critical banking infrastructure, potentially giving attackers access to systems used by multiple financial institutions.
Key findings include:
- Stolen financial services credentials surged 56.3% year over year, with 422,747 stealer logs containing banking and financial credentials shared on underground forums between September 2025 and August 2026, up from 270,444 the previous year.
- Approximately 2,000 infected machines held credentials for critical banking infrastructure, including card-issuing platforms, core banking vendors and payment systems. A single compromised credential could potentially put multiple financial institutions at risk.
- 2,644 stealer logs contained both financial platform credentials and matching corporate email addresses, indicating potential employee infections. At one major U.S. bank, nearly 40% of employee-linked infected machines held credentials for internal staff systems, creating a potential entry point for attackers.
++
Intezer released its 2026 AI SOC Report for CISOs, revealing a critical disconnect between how security teams prioritize alerts and where threats actually originate, with dozens of real threats per organization each year traced back to alerts typically classified as low-severity or informational. Key findings include:
- Nearly 1% of confirmed incidents originated from alerts initially labeled as low-severity or informational. On endpoints, that figure climbed to nearly 2%. For a typical organization generating roughly 450,000 alerts per year, that’s about 50 real threats annually, or one per week, that are never investigated.
- Fewer than 6% of malicious phishing emails contained attachments. Most relied on links, language, and abuse of legitimate services such as code sandboxes, cloud file sharing, and CAPTCHA mechanisms to evade detection.
++
Picus Security’s Blue Report 2026, based on more than 338 million attack simulations run in production environments from January to June 2026, found that improving overall prevention still leaves critical gaps. Average prevention rose from 62% to 69%, yet separate tests starting with authenticated access revealed how much attackers could still do after gaining a foothold.
Key findings include:
- The same credential-theft tool was blocked 94% of the time with one method and just 3% with another.Controls stopped 94% of Mimikatz attempts to dump credentials from LSASS memory, compared with 17% targeting RDP credentials in memory and 3% targeting LSA secrets in the Windows registry. Protection against the familiar method did not extend to other ways of stealing credentials.
- Nearly two-thirds of post-compromise actions went unblocked. Autonomous pentests found an average prevention rate of just 37% across tested attacker actions. Discovery and collection actions, including mapping domains and enumerating file shares and sessions, averaged only about 10% prevention.
- 93% of simulated data exfiltration attempts went unblocked. Controls stopped just 7%, making data exfiltration the least prevented attack vector for the fourth consecutive year.
- 86% of simulated attacks generated no alert. Logging improved from 54% to 58%, but the alert score remained at 14%, unchanged from 2025. Organizations recorded more attacker activity without improving how often that activity triggered a warning.
++
SimSpace’s State of Agentic Cybersecurity report, which combines survey findings from global security leaders with data from realistic cyber simulations, reveals a significant gap between confidence in AI-driven defenses and actual security readiness. Key findings include:
- 78% of security leaders report high confidence in their defenses, but security teams have scored as low as 30% in realistic cyber exercises.
- 73% of organizations are already using AI agents in their SOCs, but only 29% conduct security testing with cyber simulations.
- 44% of organizations test biannually or less, or not at all, despite the growing adoption of AI in security operations.
- Deploying AI tools in the SOC initially reduces performance by 10–20% before improvements are realized through repeated testing.
++
Spur’s September 2026 investigation into the Mellowtel SDK reveals how ordinary browser extensions can quietly transform consumers’ internet connections into nodes in remotely controlled web-scraping networks. Researchers uncovered software embedded in familiar utilities, including timers, calculators and screenshot tools, that could route third-party traffic through residential IP addresses while stripping website security protections and operating largely out of sight.
Key findings include:
- Approximately 90,000 IP addresses were initially associated with the Mellowtel proxy network, revealing the scale of an operation capable of routing web-scraping activity through everyday internet connections.
- Approximately 245 browser extensions, with roughly 1 million combined installations, were previously identified as containing the Mellowtel SDK, according to SecureAnnex research cited by Spur.
- 43% of observed Mellowtel-associated IP addresses were located in the United States, followed by China at 11% and Great Britain at 9%, demonstrating the network’s international reach and heavy concentration in U.S. residential connections.
++
Crogl’s State of SecOps 2026 report, based on independent research conducted by Ponemon Institute, reveals a concerning gap between the volume of security alerts organizations receive and their ability to investigate and respond to them. Key findings include:
- 4,330 alerts, seven staff members: The average organization generates 4,330 security alerts every day, with just seven IT staff members responsible for managing them.
- Nearly two-thirds of alerts go uninvestigated: Security teams investigate just 37% of the alerts they receive, leaving thousands of potential warning signs unexamined each day.
- Less than half of organizations feel equipped to respond: Only 43% of organizations rate themselves as effective or highly effective at identifying and responding to security alerts.
- Integration challenges slow AI deployment: Half of respondents (50%) cite difficulty integrating AI into existing security workflows as a top deployment barrier, followed closely by fragmented data that’s difficult to normalize (49%).
- AI alone isn’t closing the gap: More than half of respondents (52%) rate human analysts as highly effective as the final line of defense in an AI-powered SOC, compared to 44% who say the same of AI alone.
++
Here are four scary findings from NowSecure on the security risks organizations face with mobile applications.
- 97% of 136,711 tested mobile apps showed no detectable post-quantum cryptography (PQC) code, leaving many applications unprepared for future quantum computing threats. This creates concerns about attackers collecting encrypted data today that could be decrypted in the future. The report can be found here!
- 37% of respondents have not implemented AI behavior monitoring for their mobile apps, despite 95% reporting AI use in their apps. As more organizations introduce AI into their mobile apps, many still lack the visibility needed to identify potential security risks. The report can be found here!
- 24% of tested mobile apps potentially expose functionality to AI agents, raising concerns about how AI agents could access app features and potentially sensitive information without proper security controls in place. The report can be found here!
- 68% of respondents say more than half of their mobile app code comes from third parties, yet only 49% always assess that code before release. This creates potential security risks as organizations rely heavily on third-party code without consistently checking it for vulnerabilities. The report can be found here!
++
Below are five findings that show some of the security risks organizations are facing across Mac and mobile devices.
- 95% of mobile applications assessed contained at least one medium-severity vulnerability, indicating widespread security weaknesses in apps used by organizations. These vulnerabilities can create opportunities for attackers to compromise sensitive information.
- 53% of organizations had at least one mobile device running a critically outdated operating system, leaving devices exposed to known security vulnerabilities. Without timely updates, organizations risk leaving devices vulnerable to threats that could otherwise be prevented.
- 50% of Mac malware was classified as trojans, an increase of more than 33 percentage points from 2024. Trojans have become the most prevalent malware category affecting Macs, showing how attackers increasingly disguise malicious software as legitimate applications.
- 44% of Mac devices showed malicious network traffic, demonstrating how frequently devices encounter suspicious or potentially harmful network activity. This shows the importance of monitoring network connections to identify potential threats.
- 25% of organizations had at least one mobile user click on a phishing link, showing the continued threat of mobile phishing attacks. These attacks can trick employees into revealing credentials or exposing sensitive company information.
These findings come from two of Jamf’s reports:
++
Coalition’s 2026 Cyber Claims Report revealed trends in rising cyberattack frequency, escalating ransomware demands and the financial consequences of overlooked security risks:
- Ransomware demands jumped 47% year over year, reaching an average of more than $1 million in 2025. Some victims faced demands as high as $16 million.
- Business email compromise claims frequency increased 15% year over year, accounting for 31% of all claims in 2025.
- Cyber claims frequency among businesses with less than $25 million in revenue rose 10% year over year.
- 872 companies ignored critical alerts during the quoting process in 2025 and later suffered ransomware attacks totaling $436 million in losses.
++
AvePoint’s The State of AI 2026: Scaling Trust, Control, and Readiness in the Agentic Era, surveyed 750 respondents with direct responsibility for information management, data security, or AI programs. The results show a consistent pattern: AI adoption is scaling faster than enterprise governance, visibility, and operational control:
- 89.5% of organizations experienced at least one generative AI-related security breach in the past 12 months, up from 75.1% in the 2025 data set. 88.4% experienced at least one security breach due to AI agents.
- 17.6% of organizations do not know whether employees are using unsanctioned generative AI tools, nearly tripled from 6.3% in 2025. For AI agents, 21.1% are unsure whether unsanctioned tools are being used to create agents.
- 82.7% of respondents report being very or extremely confident in their ability to prevent unauthorized data access. However, among those reporting the highest confidence, 62% still experienced AI-related unauthorized access incidents, rising to 72% among those reporting “very confident.”
- 86.9% of organizations delayed generative AI deployments by an average of 5.88 months due to data security and data management concerns. Similarly, 86% delayed AI agent deployments by an average of 5.92 months.
- 35.5% of enterprise data is currently generated by generative AI assistants, and organizations expect that share to rise to 42.1% within 12 months. Meanwhile, 46.9% of employees already rely on AI agents weekly or daily.
++
SMBs are building AI faster than they can govern it, and every unmonitored agent or workflow is a potential entry point for attackers. Usage policies written for employees using off-the-shelf tools don’t reach custom agents and workflows, and usage-based pricing makes costs hard to forecast. For SMBs and the IT partners who support them, the first step is finding out what AI is running, then securing and governing it before shadow AI becomes a breach and surprise bills become a budget problem. Pax8 recently released its Q3 2026 SMB AI Pulse Report exploring how SMBs are leveraging AI, finding:
- 67% of SMBs using AI have built at least one AI solution that is now in regular use, and 46% of those builders are building or testing AI agents that complete multi-step tasks on their own
- 35% of SMBs using AI discovered shadow AI in the past six months, meaning tools, workflows, or agents running in their business without official approval.
- Only 40% of businesses have a documented policy on how employees use AI and another 35% rely on informal guidelines, but these policies generally don’t cover the agents and workflows a business builds itself, leaving them largely unchecked.
- 49% of SMBs that paid for AI in the past year were hit with a bill or usage charge higher than they expected, rising to 57% in the U.S.
- 55% of decision-makers personally spend three or more hours a week reviewing, editing, or fact-checking AI-generated work
++
Auvik’s 2026 IT Trends Report surveyed IT teams and MSPs across the U.S. and Canada on AI readiness, shadow IT, budgets, work models, and operational maturity.
Key findings include:
- 61% of respondents discover unauthorized SaaS applications at least monthly, and 23% find them weekly
- 20% of MSP respondents named shadow IT the number one issue business leaders aren’t paying enough attention to, ahead of disaster recovery and backup at 16%
- 44% say a lack of real-time visibility into their environment impedes their ability to work effectively, and 27% still cite network visibility as a top challenge
- Only 42% of help desk respondents say their organization has an AI policy and just 5% say AI is core to their IT operations today
As cyberattacks become faster and more automated, those blind spots become increasingly difficult to afford. Attackers can exploit unmanaged devices, applications, and vulnerabilities before periodic reviews uncover them. Keeping pace requires continuous, unified visibility into what is connected to the network and where exposure exists, rather than relying on a quarterly audit that may already be outdated by the time it is completed.
++
A recent survey of 202 enterprise IT and security leaders, conducted by Enterprise Management Associates (EMA) for API security and agentic AI protection company Cequence Security, found that nearly 65% of organizations have had an AI agent act outside its intended scope, with 29% reporting measurable damage. Additional findings include
- Only 32% can detect and contain an out-of-scope agent action within minutes, while 55% need hours and manual steps to shut it down.
- 3.5% first learn an agent went rogue when a customer or external partner reports it.
- 45% cannot easily produce a complete record of everything a specific agent did over the past 30 days.
- 47% cannot reliably inventory every AI agent running in their environment.
++
Hoxhunt released its 2026 Phishing Trends Report, based on more than 50 million phishing simulations and real attacks that bypassed email filters. The report found that AI-generated phishing jumped from a rare threat to roughly half of all reported attacks over the holidays.
Key findings include:
- AI-generated phishing surged 14x in December 2025, from 4% of reported phishing emails in November to 56%.
- Malicious calendar invites (.ics files) had a 24% failure rate in simulations, 4 to 6 times the global baseline.
- Malicious scalable vector graphics, or SVG attachments rose fifty-fold in 2025 compared with 2024.
- Callback phishing, which pushes victims to call a phone number instead of click a link, appeared in 4.9% of AI-assisted phishing emails.
- Users in a behavior change program cut malicious clicks by 87% and improved reporting 6x within six months.
++
Noma Labs, the research team at Noma Security, has spent 2026 disclosing vulnerabilities in widely used AI agent tools, including Ruflo, Context7, GitHub Agentic Workflows and Claude Code Desktop. Its work shows that attackers can hijack enterprise AI agents by planting instructions in content the agent already trusts. Each flaw was reported to the vendor and has since been fixed.
Scariest findings include:
- A CVSS 10 flaw in Ruflo, a popular AI agent tool, left 233 of its tools open to anyone with no password, so a single request could give an outsider control of the system. The platform has more than 67,000 GitHub stars and roughly 10 million downloads.
- In GitLost, an unauthenticated attacker posted a crafted GitHub Issue in a public repository and got GitHub’s AI agent to publish private repository contents. Adding the word “Additionally” was enough to get around the guardrails.
- In Workflow Identity Hijacking, a single unauthenticated email to a company’s support inbox was enough to get the quarterly sales numbers back within minutes. No prompt injection or stolen credentials were needed, because the AI workflow ran on high-privilege accounts and never checked who was asking. Noma found the flaw in Google Workflows, and Google confirmed a fix.
- The SneakPeek exploit in Claude Code Desktop was scored CVSS 8.7 by Anthropic. A poisoned file over 256KB could quietly steal a user’s private login keys, cloud credentials and access tokens from Claude Code Desktop when they simply asked Claude to review it. A first patch was bypassed before a full fix shipped on August 17, 2026.
++
Black Duck’s The State of AI-Powered Software Development report, based on a UserEvidence survey of 831 software engineers and DevOps professionals, observed that 97% of respondents actively use AI coding assistants. Productivity is up, but governance and security testing haven’t kept pace, and the work is shifting downstream rather than disappearing.
- Only 30% of teams have fully governed AI coding assistant use, meaning tools are formally approved, centrally managed and actively monitored. A quarter have no defined AI coding policies at all.
- 90% of teams run into problems with AI-generated code, and the biggest slowdowns are manual review (52%) and security testing (51%).
- Many teams can’t reliably tell which code AI wrote. 38% depend on developers to note AI use manually in pull requests, and 7% don’t track it at all.
- Executives and developers see AI code quality very differently. 48% of C-suite respondents rate it excellent, compared with just 8% of technical contributors.
++
SailPoint’s 2026-2027 “Horizons of Identity Security” report, based on a survey of 340 identity, IT, cybersecurity, and risk leaders. It found that 79% of organizations run AI agents in production, but only 2% use identity security tools built to govern them, a 40-to-1 gap.
Additional key findings include:
- 57% of leaders are confident they can meet AI-related regulatory requirements, but only 43% feel prepared to produce verifiable evidence in an AI audit.
- 69% already include AI agents in identity audits, and 81% expect audit workload to grow over the next 12-24 months.
- Financial services are among the most mature industries for identity security, alongside technology, yet 54% of financial services organizations remain at the two lowest maturity tiers, and only 10% have reached the advanced tiers.
- AI agents now represent 22% of all non-human accounts, yet 85% of organizations rely on legacy general-purpose identity tools or platform-native controls.
++
Qualys Threat Research Unit (TRU) research on remediation, based on more than one billion CISA Known Exploited Vulnerabilities (KEV) remediation records from over 10,000 organizations, found that teams are closing far more vulnerabilities while falling behind on the most dangerous ones. A TRU analysis of the PHP2Shell flaws puts a number on how much PHP is exposed. Scary stats include:
- Closed vulnerabilities rose from 73 million in 2022 to 473 million in 2025, a 6x increase, yet the share of critical vulnerabilities still open at day seven rose from 56% to 63%.
- Across the most critical, actively weaponized vulnerabilities of the past four years, manual remediation failed to keep pace with attackers 88% of the time, and half were weaponized before a patch existed.
- 15% of organizations that operationalized their remediation pipeline patched by the point a vulnerability was added to the KEV list.
- Of the 156 million PHP installations with a known version, 110 million (71%) run end-of-life or end-of-support versions, and most will never receive a vendor patch.
- Only 581,000 of those installations (0.37%) were patched in the first 24 hours after disclosure.
++
Ontinue’s 1H 2026 Threat Intelligence Report draws on observations from customer environments, incident investigations, threat intelligence research, and data from Ontinue’s Advanced Threat Operations team.
Key findings include:
- Approximately 30% of monitored organizations experienced at least one Microsoft Teams IT-support impersonation attempt.
- Approximately 1 in every 990 external Teams contacts showed signs of impersonation activity.
- Ontinue identified 10 confirmed device-code phishing compromises across five customer tenants in a two-month period.
- More than 25% of confirmed ClickFIx findings leveraged blockchain-based command-and-control infrastructure.
- Researchers observed 6,188 delivery domains tied to just 158 payload hosts, highlighting the increasing industrialization of attacker infrastructure.
++
Acceldata’s research with GLG found that half of Databricks users spend $5 million or more a year and still rank Databricks Unit (DBU) cost predictability as a top pain point. In contrast, Snowflake users point to compute cost volatility and poor cost governance. When 98% of executives say a 30% savings would be enough to rip out the platform their business runs on, loyalty has nothing to do with it. Enterprises can’t see what they’re paying for, and the bill keeps growing in the dark. – Mahesh Kumar, Acceldata’s CMO.
++
According to Parloa’s State of Agentic CX report, only 1% of enterprises are structurally prepared for agent-to-agent interactions. The report also found that just 8.9% of chatbot interactions resulted in resolution and 99% of voice systems relied on decades-old automation or none at all. As personal AI agents increasingly make purchases, book reservations and negotiate on consumers’ behalf, businesses will need to fix broken customer experiences today while building the infrastructure to securely interact with AI agents tomorrow. (Report)
++
“According to DigiCert, although 87% of organizations report they are planning, testing or implementing post-quantum cryptography (PQC) initiatives, deployment has increased by just two percentage points since last year’s survey. As a result, only 7% of organizations have deployed quantum-safe or hybrid cryptography across most of their digital certificates, underscoring how much work remains to transition from planning to implementation.” Source – DigiCert
++
According to SHL’s Talent Mobility at Work in 2026 report, 83% of workers would retrain for a more senior role, yet 66% believe they have to leave their employer to advance their career. The survey of 2,036 U.S. and U.K. workers also found that only 42% say their manager fully understands their skills beyond their current role, and 59% of U.S. workers say their employer favors outside candidates over developing its own people. With 89% willing to take a formal skills assessment, employers looking for talent may already have it on the payroll.
++
According to a survey of 1,000 U.S. adults commissioned by The XI Code, 60% say society does not address burnout or emotional well-being, and nearly 40% feel emotionally exhausted several times a week. The survey also found that 72.6% feel mentally overstimulated even when they’re trying to rest, and financial stress is a top driver of burnout for nearly half of respondents. 44.4% have turned to AI-based wellness tools, and 18.6% have felt more emotionally connected to an AI conversation than to a recent one with a real person. Still, 67.8% would turn to a human first during intense emotional distress. With more than 45% open to wellness solutions based on emerging science, people are looking for support that goes deeper than another app.
++
According to Lyric’s new white paper, The State of Healthcare Decision Intelligence in Payment Accuracy, health plans should measure AI maturity in payment integrity by how well each function coordinates decisions, monitors outcomes and improves over time. The framework, which draws on more than 35 years of work with health plans covering 200 million lives, finds that pre-payment editing is the most advanced function and subrogation still relies most on human judgment. As AI takes on more claims work, health plans will need to be explicit about who holds which decision rights, so they can automate more without losing accountability.
++
According to Greyparrot’s 2026 Aluminum Report, the U.S. landfilled 61 billion of the 107 billion aluminum cans it shipped in 2023, a 43% recycling rate and the lowest in decades. Those discarded cans were worth $1.2 billion and represent about 11 million megawatt-hours of energy a year, enough to power more than 1 million homes. Meanwhile, net imports covered 60% of U.S. aluminum supply in 2025. AI camera data from Brooklyn’s Sunset Park recycling facility shows that 89% of aluminum is recoverable once it reaches the sorting line, about 91,000 cans a day. As the tech sector tallies every megawatt-hour going into AI, the energy buried in landfills is a domestic supply the U.S. is leaving uncounted.
++
According to eHealth’s Medicare Annual Enrollment Period Consumer Trends survey, 71% of Medicare beneficiaries say shopping for coverage is confusing, and 48% skipped or delayed care in the past year because they were confused about their coverage. The survey of more than 1,000 beneficiaries also found that 55% of Medicare Advantage and Part D enrollees wrongly believe their plans cover GLP-1 drugs for weight loss, and 59% of Medicare Advantage enrollees would switch plans to get that coverage, even if it meant losing their preferred doctors. Nearly half have already used AI tools to review plans. Still, 61% would trust a licensed agent over AI when the two disagree, which makes expert guidance more important this enrollment season, as 28% say they’ve already fallen victim to a Medicare scam.
++
According to Fountain’s Louder Than Words report, only about 8% of U.S. frontline applicants are hired, and 881,000 applications timed out in 2025 before anyone made a decision. The analysis of 12.5 million frontline applications also found that just 19% of applicants reach an interview, and a quarter of hires wait six weeks or more for an offer, even though more than half of frontline workers couldn’t cover a month without income. AI now runs 14% of first interviews, up from 2% a year earlier, so employers that want to keep hourly applicants interested need to answer them faster.
++
“Globalgig’s Future of AI-Driven Networks 2026 report reveals a striking confidence gap: while 85% of enterprises believe they have the in-house expertise to manage autonomous AI networks, 67% admit their biggest fear is deploying AI without the proper expertise. More than half (51.8%) are already using AI for security threat detection, yet concerns about data privacy and the risk of AI making incorrect security decisions remain top of mind.” (Press Release)
++
“Game industry layoffs in 2026 passed the 2025 total by late July. By July 26, one industry tracker had counted 9,781 confirmed job cuts, against 9,197 for all of 2025, and projected 14,259 for the full year, just below the 2024 record of 15,631.”
“Studios are running live games with smaller teams while mobile downloads keep falling. AI agents can take on the legwork of pulling data, checking it and explaining what changed, so a producer can ask a question in the morning and act on the answer the same day. The team still makes the decision, using data that is still current.” – Chris Han, co-founder and president, ThinkingAI
++
According to a new EQUS survey of U.S. adults who use AI assistants, 65% have told their assistant about someone else. Among full-time workers, 1 in 4 have shared details about a coworker or client. Sandy Carter, CEO of EQUS shared: “Ownership has a simple test. Can you see what your AI knows about you, correct it, take it with you and take it back? If the answer is no, you don’t own it. You’re renting it.” (Press release)
++
Here are some scary cybersecurity statistics from Securonix’s “The Evolution of Cybersecurity Automation and AI Adoption” report:
- 19% say AI runs alert triage without human approval, while 35% say it will always need human oversight. Some organizations are giving AI control over a task many still consider too consequential to automate fully.
- 32% say their AI governance is ad hoc or emerging. Nearly one-third are adopting AI while their governance practices are still developing.
++
Zimperium’s 2026 Global Mobile Threat Report, drawn from its zLabs threat research and analysis of millions of enterprise mobile devices and apps, found that AI is speeding up mobile attacks faster than enterprises can prepare. Phishing is the most visible shift, but AI is also changing how malware is built, how employees use apps, and how mobile apps themselves are developed.
- Phishing on mobile is surging: Phishing events detected on employee devices are up 380% since January 2025, and the number of devices where an employee clicked a malicious link rose 110% in 2025 over the prior year. Nearly 86% of phishing attacks now contain AI-generated elements.
- Spyware is spreading: It’s now present on nearly one in ten mobile devices, more than 4x the prior year.
- Shadow AI is growing inside mobile apps: AI adoption in mobile apps has grown 14x on Android and 7x on iOS, creating new blind spots for security and compliance teams around data privacy and IP protection.
- Attackers are using AI at scale: The median threat actor now uses AI across 15 MITRE ATT&CK techniques, and some use 40 to 50.
- Vibe-coded apps add risk: The report projects that 25% of software defects will trace to AI-generated code by 2027, up from less than 1% in 2023.
++
Below are five findings from Cynet’s 1H Echo Report on the growing risks organizations face from credential theft and ransomware attacks.
- 8 in 10 host breaches investigated by Cynet in the first half of 2026 began with stolen identities, showing how attackers are increasingly gaining access through legitimate credentials rather than breaking through security controls. Once inside, attackers can move through systems while appearing to be authorized users.
- Akira ransomware attackers achieved full encryption in under an hour in some cases investigated by Cynet, leaving organizations very little time to detect and stop an attack before their systems are locked.
- The Gentlemen ransomware group claimed 467 victims in the first half of 2026, but leaked internal data suggested the actual victim count could be nearly four times higher. This suggests that publicly reported ransomware activity may significantly underestimate the true number of affected organizations.
- The Qilin ransomware group recorded 648 claimed leak-site postings in the first half of 2026, demonstrating the continued scale of ransomware activity. These postings show how frequently attackers publicly claim victims to pressure organizations into paying.
- Roughly 1 in 3 host breaches investigated by Cynet involved an SSL-VPN gateway, with attackers typically using stolen passwords rather than exploiting vulnerabilities in the gateway itself. This shows how compromised credentials can turn trusted remote access tools into entry points for attackers.
++
HackerOne, 10th annual Security Research Report
The report found that the backlog of validated, unresolved vulnerabilities grew 131% over two years, even as average resolution time decreased 54%, from 135 to 62 days. HackerOne calls this growing gap between discovery and remediation “Exposure Debt.”
A few additional findings worth highlighting:
- Nearly 70% of security leaders say validated findings are arriving faster than their teams can fix them.
- AI-related vulnerability reports increased 53% year over year, with system prompt leakage reports up 557%.
- 85% of security researchers are actively developing their AI skills, and 68% say AI has shifted their focus toward more complex, higher-bounty vulnerabilities.
++
Radware, H1 2026 Global Threat Analysis Report
Web DDoS attacks more than doubled in the first half of 2026, rising 110.6% year over year. In just six months, Radware mitigated nearly 83% of the web DDoS attacks it handled during all of 2025. Malicious web application and API transactions also increased 104%, averaging more than 14,000 per application each day.
Radware / Osterman Research, AI Agent Security Survey
An Osterman Research survey of 377 organizations commissioned by Radware found that 77% are using or implementing AI agents or autonomous workflows, while only 17.2% have complete visibility into their AI agents and AI-driven processes. API visibility is similarly limited: 81.1% update production APIs at least weekly, but just 6.9% fully document all of their APIs.
++
Lineaje AI Skills Marketplace
AI skills are rapidly becoming the primary execution layer of enterprise AI, enabling agents to automate workflows, access systems, and act autonomously at scale. Lineaje’s AI Threat Labs research found that of 52,755 skills analyzed across major marketplaces, 2.1% were flagged as unsafe — and of those, 62.9% were confirmed malicious, with the remaining 37.1% exhibiting suspicious behavior.
While 2.1% may appear negligible, it is exceptionally high by malware standards and the stakes are amplified by the fact that skills routinely operate with the same privileges as human users.
++
Portnox / IDC InfoBrief
An IDC InfoBrief, commissioned by Portnox, shows what enterprise buyers are prioritizing in 2026. The research finds 18% of respondents ranked agent-to-agent access as a priority among access control types in 2026 zero trust plans and identified network access control (NAC) as the second most important technology for the visibility and context security teams need. It also warns that spotting risky AI access is not enough if enforcement takes minutes while exploits take seconds. 41.6% said security and compliance requirements were an important factor in deploying AI extensively.
++
Here’s the scary stat: organizations are averaging six insider-driven incidents a month…at about $13.1 million each. That’s roughly $79 million a month, and it’s a recurring bill, not a one-off.
Intent doesn’t tell you much anymore. Malicious, negligent, compromised… the exposure is the same. According to Mimecast’s State of Human Risk 2026:
- It’s balanced: 42% saw malicious insider incidents rise, and 42% saw negligent ones rise.
- The pieces aren’t connected: Only 28% pair security awareness training with continuous monitoring.
- It’s left the inbox: 45% report more attacks on collaboration tools, and 64% say native controls aren’t up to the task.
- AI is the accelerant: 80% worry about sensitive data leaking through GenAI tools.
And that worry is warranted: separate Mimecast research found 98% of organizations have unsanctioned AI tools active… mostly employees just trying to get work done.
Bottom line: give people sanctioned tools, and put adaptive controls where risk shows up: email, Slack and Teams, and AI workflows. Visibility into what people do with your data is the difference between security and hope.
++
Kai‘s 2026 State of Autonomous Defense – Released in July 2026, this is a global survey of 500 CISOs, examining how organizations are adapting to machine-speed threats, and what separates the ones already moving.
Key findings:
- 63% of CISOs believe attackers currently have the advantage because of AI.
- 48% say at least one-quarter of known vulnerabilities remain unremediated for more than 30 days.
- 77% report that vulnerability management contributes at least moderately to security team burnout.
- While just 35% of organizations describe their vulnerability management approach as mostly or primarily machine-led today, that figure is expected to reach 45% within the next 12-18 months.
++
Key finding:
- An AI agent only needs $12 and eight minutes to reach full domain control by chaining together credentials, misconfigurations and trust relationships.
++
Mitel’s 2026 State of Workforce Communication Report– A global survey of 2,000 IT leaders, desk, and frontline workers and workers across North America and Europe were asked what’s working and what isn’t in workforce communications. Based on research by Vanson Bourne, it reveals a significant gap in today’s era of widespread AI adoption: while workplace communication is mission-critical, tools are misaligned with how teams execute, forcing employees to quietly compensate at measurable cost to productivity, security, and service quality.
Key finding:
- 76% use non-approved communications channels, increasing risks around data exposure, compliance, cybersecurity, and visibility/control, according to 90% of ITDMs.
++
2026 Pindrop® Deepfake Readiness Index – Report that examines the growing threat of deepfake attacks and enterprise readiness to defend against them
Key findings:
- 93% of security leaders say they are concerned their organizations aren’t prepared for a deepfake attack
- 74% of organizations surveyed have encountered or suspect a deepfake attack in the past year, yet only 10% report having purpose-built tools to address the threat
- 1 in 3 enterprises are still relying on legacy defenses to combat deepfakes
- Only 10% have invested in purpose-built tools
- Among organizations that experienced or suspected a deepfake attack, nearly half reported costs of $500,000 or more, including direct losses, remediation and staff time.
- One in four reported costs of $1 million or more from a single incident.
- 49% of affected organizations faced a follow-on cyberattack, including ransomware
- 37% think a single deepfake attack could put a company like theirs out of business
++
Saviynt‘s 2026 CISO AI Risk Report – Report on the real state of AI identity risk inside large enterprises, including unsanctioned AI tools, uncontrolled agent privileges, and the fastest-growing governance blind spot CISOs now face.
Key findings include:
- 92% of organizations lack full visibility into AI identities, and 95% doubt they could detect misuse if it happened.
- 75% have discovered unsanctioned AI tools currently running in their environments
- 86% lack or don’t enforce access policies for AI Identities
- 71% say AI tools already access core systems, yet only 16% say that access is effectively governed.
- 86% of security leaders lack or don’t enforce access policies for AI identities. Only 17% govern even half of their AI identities like human users, and just 5% feel confident they could contain a compromised agent.
++
Sonatype, the company helping enterprises build with confidence in the AI era, recently released its Q3 2026 Open Source Malware Index, which identified 149,329 malicious open source packages across ecosystems, bringing the total logged since tracking began in 2017 to 1,960,846.
Key findings from the Q3 2026 Open Source Malware Index include:
- Among packages containing overt-malware behavior, 74.5% involved droppers, secrets exfiltration, or both, showing how often Q3 malware was designed to extend the attack beyond initial execution.
- npm once again accounted for the overwhelming majority of packages this quarter, with 133,579 packages making up 89.5% of the quarter’s total.
- 82.7% of Q3 malware involved repository abuse.
Mini Shai-Hulud affected 2,225 affected component versions, which turned trusted access into a repeatable propagation mechanism, allowing one compromise to feed the next at scale.
++
Entrust’s KYE & Workforce Identity Crisis Report – Report that examines how AI and remote work have reshaped workforce identity, the cyber risks and visibility gaps that come with it, and what the future of organizational security needs to look like to protect sensitive data from AI impersonation attacks.
Key findings from the survey of more than 1,000 IT and security leaders in the U.S. and UK include:
- 96% of respondents consider insider risk significant to their organization.
- 91% of organizations have experienced an insider threat incident/near miss in the past two years.
- 46% cite phishing/social engineering as a top workforce identity concern, followed by credential compromise (43%), privileged access misuse (34%) and account takeover/impersonation (33%).
- Only 39% of respondents are very confident verifying identity during hiring; 43% during onboarding; 37% during account recovery; 32% for remote/hybrid access; and 31% for contractor/partner/temp access.
- Fully remote organizations reported 37% rates of employee impersonation, insider threats and candidate fraud — the highest levels cited in the report.
- 45% of respondents still rely on usernames/passwords to authenticate employees, and 28% rely on manual IT/HR verification.
- Only 4% say they have no plans to measure AI-enabled impersonation; 21% recognize the risk but don’t currently measure its impact.
++
Vectra AI – 2026 State of Threat Exposure Management Report – Released in July 2026, the report analyzes anonymized network telemetry across diverse enterprise environments, revealing how unmanaged devices, AI agents and persistent security weaknesses are creating new exposure risks and blind spots for security teams.
Key findings include:
- 98% of analyzed environments had at least one attacker-relevant exposure condition present, highlighting how widespread potential attack paths remain across enterprise networks.
- 85% of environments had plaintext passwords observed, creating opportunities for credential theft, privilege escalation and lateral movement.
- More than 30% of devices across analyzed environments were unmanaged on average because endpoint security agents could not be deployed, leaving significant portions of enterprise networks outside traditional endpoint visibility.
- 35% of environments with observed AI-agent activity had more AI agents than devices, with one extreme case reaching 96 AI agents for a single device, introducing new access and communication paths that security teams must monitor.
- 63% of analyzed environments showed exposure across multiple risk categories, including legacy protocols, weak cryptography, credential exposure and exposed remote access.
++
Lineos (an insightsoftware brand) – 2026 State of AI in Finance Report: The Insight Gap in Corporate Finance – Released in September 2026, this report surveyed 854 finance professionals at organizations with more than 500 employees, examining the growing disconnect between AI adoption, trust in AI-generated insights and the challenges of outdated financial systems.
Key findings include:
- 60% of finance professionals say they always or often manually verify AI-generated insights before acting on them, highlighting a persistent trust gap that threatens to erase AI’s promised time savings.
- Only 36% of finance professionals completely or mostly trust AI-generated financial results, raising concerns about the reliability of AI-driven decision-making.
- 20% of finance professionals worry AI will replace their jobs, up seven percentage points year over year, making job displacement fears the single biggest barrier to AI adoption.
++
Imprivata – The Agentic AI Trust Gap: Why Healthcare Needs Identity-Led Governance – Released September 2026, this report explores emerging security risks as AI agents take on a larger role in patient care and healthcare operations, including how organizations are assigning identities and access to agents, where human judgment remains essential, and what safeguards are needed before agents can act independently.
Key findings from 250 U.S. healthcare leaders responsible for identity and/or AI strategy include:
- 83% of organizations have already deployed AI across multiple departments or use cases, yet 72% report AI tools or agents are deployed without formal IT approval, highlighting a clear challenge with “shadow AI.”
- 73% of organizations with a Chief AI Officer overseeing AI strategy still report AI tools being deployed without formal IT or security approval.
- 88% of respondents expect AI agents to operate with at least some degree of autonomy across clinical and operational workflows.
- Only 17% believe existing identity approaches are sufficient without adaptation.
++
Frontline Friction: The State of Authentication for Frontline Workers – Released October 2026, this report examines how authentication friction affects frontline workers in production, delivery, and other shift-based, shared-device environments.
Key findings fro more than 700 global IT and security decision-makers include:
- 34% (rising to 40% in healthcare) report authentication difficulties delaying patient care, production, delivery and/or other frontline work.
- 95% of organizations have observed frontline workers using at least one workaround because of authentication difficulties, including staying logged in to avoid re-authentication (40%), attempting to bypass MFA or other controls (36%) and sharing credentials (33%).
- 62% report a confirmed or suspected unauthorized account recovery, password reset or unlock attempt in the past year. Organizations also report confirmed or suspected helpdesk social engineering (60%) and AI-assisted/deepfake employee impersonation (57%)
- 67% of organizations remain password-based or inconsistent, while just 14% are passwordless-led.
++
LevelBlue – Q3 Tactics, Techniques, and Procedures (TTP) report – Released October 2026, this report explore the latest trends, techniques, and procedures defenders are actively facing, based on frontline threat intelligence from LevelBlue’s global digital forensics and incident response (DFIR) investigations during Q3 2026.
Key findings include:
- Data exfiltration rose from 31% in Q2 to 40% in Q3, while simply viewing/accessing files fell from 67% to 38%
- Exfiltration/extortion-only ransomware increased from 23% to 33%, while encryption-only fell from 23% to 17%
- Phishing remained the leading initial intrusion vector but dropped from 65% to 42%
- Exploited vulnerabilities more than tripled as an initial intrusion vector, from 4% to 14%
- 56% of engagements began within 10 days of compromise, up from 49% in Q2
- Technology and software became the most targeted industry at 14%, up from 5% in Q2
++
OutSystems – Engineering Business Outcomes: The Shift to Agentic Systems: Released in October 2026, OutSystems and KPMG partnered with CIO Dive to survey 858 executives across North America, Europe, Japan, India, and Australia on how they are deploying and governing agentic AI. The survey was conducted in June 2026.
Key findings:
- Only 5% of enterprises had automated controls, audit trails and human review points for agentic systems.
- 67% of respondents considered security, permissions and policy enforcement across agents a leading orchestration hurdle.
- Only 26% of leaders said their organizations were “very well prepared” to resolve conflicts between AI agents, despite 82% believing their governance could keep pace with adoption.
- 72% of organizations were operating more than 20 distinct AI tools, models or platforms.
- 36% of organizations had a fragmented approach to deciding where and how to use agentic AI.
- 38% cited reliability across chained workflows as a top orchestration hurdle.
- 35% cited data consistency and context sharing as a major orchestration hurdle.
- 31% had limited post-deployment visibility into agentic AI.
++
Doppel – Faking a brand is easy. Making it stop is hard – Released in August 2026, Doppel partnered with a16z on a report examining the growing threat of brand impersonation and how these attacks are evolving as AI advances. The report analyzed confirmed impersonation activity across hundreds of brands, with the findings below drawn from Doppel’s threat intelligence data.
Key findings:
- 94% of brands catch fewer than half of their own impersonators and the median brand catches none.
- 80.6% of impersonated brands are targeted across two or more surfaces at once.
- Multi-surface impersonation campaigns have grown ~7x in just two years.
- 91% of confirmed fakes are found by Doppel before the brand discovers them.
Doppel Threat Intelligence: An Inside-Out View of Japan’s Phishing Problem – Released in July 2026, this report highlights key findings from Doppel’s five-week honeypot project in Japan, revealing how localized threat actors operate, including attacker timing patterns, the most impersonated brands, and common tactics.
Key findings:
- 40.1% of traffic captured by the Japan honeypot was a genuine malicious lure designed to defraud recipients.
- 4 in 5 impersonation campaigns observed in Japan mimicked technology brands.
- 67.8% of malicious lures observed in Japan arrived on Mondays or Fridays.
- Financial brands are the most frequently impersonated individual brands with credit card and consumer finance companies leading the top 10, followed by e-commerce and global card/payment networks.
- 51.5% of activity captured in Japan was attacker reconnaissance – including attempts to determine whether servers were live or could be abused.
++
Ping Identity – Trust AI to Act? Consumers Draw the Line. – Released October 2026, this report surveyed 11,000 consumers across 12 countries about how much responsibility they’re willing to hand over to AI, and what that means for organizations as AI begins acting on their behalf.
Key findings:
- Only 12% of consumers fully trust organizations to manage their identity data, down from 17% last year.
- 71% of consumers are concerned an AI agent acting on their behalf could misrepresent their personality, ethics or personal values.
- 67% would not give an AI agent read access to their primary email inbox, even if doing so could save them five hours of admin work per week.
- 20% of consumers would immediately leave a website if they suspected undisclosed AI, and another 20% would avoid that brand or company in the future.
++
ServiceTitan – 2026 State of AI in the Trades Report – surveyed over 1,000 residential and commercial trades contractors, revealing what these professionals want from industry-specific AI, what’s currently holding back adoption, and how workforce shortages are shaping where they deploy the technology.
Key findings:
- AI’s productivity gains are outpacing business results: Among AI users, 64% report productivity gains and 55% report faster decision-making. At the same time, 47% report improved profitability and 43% report increased revenue, highlighting a persisting gap in translating productivity into broader business outcomes.
- Big tech thinks AI will replace workers, but the trades tell a different story: Difficulty hiring and the workforce gap are the top reasons contractors cite for experimenting with AI (37%), followed by providing off-hours support (34%) and reducing operating costs (33%), showing how AI is helping contractors support operational needs amid persistent workforce pressures.
- Barriers have shifted from cost to understanding: High costs ranked last among the key barriers, and instead, 44% of contractors cited a lack of understanding of how to use AI in their business as their main challenge. This suggests today’s contractors are more focused on successfully implementing AI and generating meaningful value from the technology than on just paying for it.
- Refining vertical AI is the next battleground: 38% of contractors say AI trained on trade-specific data would encourage greater adoption, underscoring the growing demand for technology tailored to the workflows and operational needs of the trades.
++
Chainguard – The AI Exploit Gap – Released in October 2026, this global survey of 1,300 cybersecurity leaders examines how AI is changing the security landscape.
Key Stats:
- 91% of security leaders say AI-accelerated attacks have changed their job more than anything else in the past decade.
- 77% believe it’s likely an attacker will exploit a software vulnerability at their organization in the next 12 months.
- 54% say their companies knowingly leave high- or critical-severity vulnerabilities unpatched at least sometimes.
++
Smartsheet – AI Is Making Work Denser, Not Lighter– Released in May 2026, this analysis of 1.4 million active enterprise projects across the Smartsheet platform examines how AI is packing more activity into the same working hours while the capacity to review, decide and approve stays the same.
Key findings:
- Automation intensity per enterprise account has grown 55% year over year, and overall activity has increased 46%.
- AI-powered document and content generation has increased 230% since 2023.
- 80% of AI-generated content is edited before it’s finalized.
- Activity has grown from 538 million actions in 2024 to 786 million in 2025, with 33.7 billion automated actions processed in the past year alone.
++
Gong Labs analyzed more than 120 million sales calls to understand who is actually participating in buyer conversations and what that means for deal success.
Key Findings:
- 4 in 10 buyers never say a word on sales calls, and higher rates of silence are linked to more crowded meetings
- Win rates increase by 30% when four or more people speak on a call
- C-suite attendees are among the quietest: 11% stay silent, compared with just 5% of VPs
- Silence varies significantly by country: Japan had the highest silence ratio at 16.5%, while Northwestern European and Nordic countries ranged from 6.5% to 8%.
- More attendees don’t necessarily mean broader buy-in: a crowded call dominated by one champion can still leave sellers with only one real perspective inside the account.
++
Scary stats from Cayosoft’s The Foundation of Identity Resilience report.
- Recovery is a priority, but preparedness remains untested: A Cayosoft study of senior IT leaders found that recovery, rollback, and resilience ranked among the top areas expected to receive attention in 2027. Yet 72% of respondents said their organizations do not conduct live Active Directory disaster-recovery drills to restore authentication and critical identity services, or were unsure whether they do. Without those drills, teams may rely on recovery procedures that haven’t been tested under post-attack conditions.
- More than a quarter of executives lack a clear recovery timeline: Over a quarter (25.8%) of respondents said their organizations had no defined recovery time expectation or were unsure of the expectations if Active Directory or core identity services become unavailable. A security incident could leave them facing prolonged disruption without a clear benchmark for when critical services should return.
- Cloud identity recovery raises unanswered questions: When asked about Microsoft Entra ID recovery readiness, nearly 22% of respondents were unclear about what they could recover, another 22% lacked clarity about a trusted restore point, and about 19% had concerns about recovery speed. Teams that cannot identify a trusted restore point risk restoring compromised accounts or configurations alongside the services they are trying to recover.
++
Protegrity / AI Friction Tax
An Enterprise Management Associates survey of 152 IT and security leaders commissioned by Protegrity found that 82.9% of organizations have experienced AI production delays due to security or compliance reviews. Among those experiencing delays, 66.7% reported setbacks of one month or longer. Security concerns are also limiting AI capabilities, with 81.6% of organizations reporting that they have deployed AI agents in a diminished form.
Additional findings include:
- 62.5% of respondents identified data leakage through autonomous agent actions as their top security concern.
- 47.4% identified reducing manual audit and compliance costs as their top desired business outcome from improved AI data security and governance.
- 91.5% said they would find a solution that embeds data protection directly into AI workflows extremely or very valuable.
++
Sophos – 2026 MSP Perspectives Report: Cybersecurity Leadership, Compliance, and Scale
- Managed Service Providers (MSPs) estimate that almost half of their customers (46%) look to them for CISO-level leadership
- 84% of MSPs expect demand for CISO services to increase over the next 12 months.
- 53% of MSPs use multiple tools or platforms to centrally manage cybersecurity compliance or CISO-type activities. Just 31% can fully automate reports at speed.
Rapid7 – Quarterly Threat Landscape Report
- Attackers are rapidly weaponizing vulnerabilities, exploiting new CVEs in an average of just 3 days. This speed is fueled by the fact that 79% of the quarter’s exploited flaws were remote, unauthenticated “holy grail” vulnerabilities.
- With 71% of all active exploitation in Q3 occurring within the first 7 days, traditional weekly or monthly patching cycles are fundamentally too slow for defense.
- Only 8% of Rapid7 tracked incidents featured AI as an operational component, with humans still driving attack intent and selecting targets.
Yubico – Global State of Authentication Report
- 43% rely on passwords at work, despite many knowing passkeys are more secure.
- 39% of security pros failed to distinguish AI-generated text from human writing.
- 44% of orgs have experienced at least one AI-driven phishing attack.
- 29% report being personally targeted by deepfake communications.
- 68% of Gen Z professionals feel comfortable permitting an AI agent to communicate with clients or colleagues on their behalf, compared to just 27% of Baby Boomers.
Horizon3 – Honeyquest for LLMs: Rethinking Cyber Deception for AI Attackers
- AI attackers fell for traps 2.1x more often than human red-teamers (78.5% vs. 37%), across 21 AI models from 10 providers.
- 73% of the time, a model that identified a trap in its own reasoning attacked it anyway.
- AI attackers took conspicuous credential and key bait 83% to 100% of the time, compared with 46% to 66% for humans.
- Decoys cut real-asset targeting by 22% for human attackers, but had no statistically significant effect on AI attackers.
Outtake – 2026 State of Digital Risk Report
- 84% of organizations experienced a material digital risk incident in the past year, yet only 7% describe their program as “leading” on a maturity scale
- 96% have no automated way to stop a hijacked AI agent, and only 4% have both automated detection and containment
- 53% had an executive or employee impersonated, but 43% have no capability to build a threat profile around someone being targeted
- 44% say AI-generated attacks are already indistinguishable from legitimate activity
##
