Industry executives and experts share their predictions for 2026. Read them in this 18th annual VMblog.com series exclusive.
By Charles Henderson, EVP and Head of DivisionHex; Cory Marchand, Director of Threat Hunting; Neil “Grifter” Wyler, VP of Defensive Services, Coalfire
The “AI Everything” era will fade as soon as it rose, but Pandora’s Box can’t be shut. Organizations will need to keep close tabs on where AI makes sense and where it doesn’t – for the sake of security, overhead costs and actually augmenting human processes for the better. Coalfire asked three of its experts to predict what security teams will need to keep top of mind for 2026, and their responses should help teams and organizations be prepared for the coming year.
Incident Response Playbooks Go Through the Shredder
Traditional IR is built for static environments. You isolate, image, rebuild, and move on. AI-heavy environments break that model.
In 2026, responders will have to deal with models that have learned from poisoned data, agents that have been subtly re-tasked, and pipelines that regenerate bad behavior every time they restart. There is no “gold image” when the system is constantly rewriting itself.
IR playbooks will go from “runbook” to “research project.” Every major AI incident will require bespoke investigation:
What did the model learn?
Where did the prompts come from?
What outputs were trusted downstream?
Containment will not just mean kicking the attacker out. It will mean unlearning everything they taught your system. – Charles Henderson
AI-Free Zones Become The New Gold Standard Of Security”
By the end of 2026, the most mature organizations in the world will flex not about how much AI they use, but about where they refuse to use it.
After a series of high-profile AI-driven incidents – poisoned models silently shaping decisions, compromised agents auto-approving bad transactions, “smart” workflows amplifying a breach instead of containing it – boards will realize something brutal:
The riskiest systems in the building are not the old legacy apps.
They are the shiny AI deployments nobody fully understands. – Charles Henderson
The “AI For Everything” Era Crashes Hard
2026 is the year security teams stage an intervention on their own AI roadmaps.
The attack surface created by half-baked models, bolted-on agents, and “smart” workflows will finally trigger enough outages, near misses, and headlines that boards ask a brutal question: what problem were we actually solving here?
Organizations will start ripping out AI that never should have existed in the first place. If a bash script, a simple automation rule, or a human with a checklist can do the job, that shiny AI pilot is going on the chopping block.
The new test is simple:
If you cannot explain the risk model, the failure modes, and the security controls for your AI in one whiteboard session, you do not have a product. You have a liability. – Charles Henderson
Cyber Villains Prioritize PII and Real-World Disruption
Attackers will double down on two categories:
- Places drowning in PII.
- Places where a digital hit creates physical pain.
Healthcare checks both boxes. Patient data is a permanent identity blueprint, and the sector still struggles with fragile, legacy infrastructure. Ransomware crews and data brokers know this. In 2026, hospitals will feel less like care facilities and more like treasure vaults under siege.
Manufacturing is where things get kinetic. You do not have to take a country offline to cause damage. Stopping a single refinery, battery plant, or port at the right time can ripple through supply chains, economies, and headlines. Expect more “mysterious outages” that look a lot less mysterious once you follow the packet trail. – Charles Henderson
Planet-Scale Outages Become Background Noise
A large-scale outage may seem like a “black swan” to most people. It’s not. It’s a dress rehearsal for what’s coming.
We’ve built a monoculture on top of a handful of hyperscalers, SaaS platforms, and CDNs. When one of them sneezes, the global economy catches pneumonia. In 2026, major outages will not be shocking events. They will be recurring line items in your risk register.
The real story is who actually treats that risk as real. The winners will be the few organizations that design for “cloud snow days” with multi-region, multi-provider, or even non-cloud contingency plans. Everyone else will be writing apologetic status updates while their “five nines” marketing slides age badly. – Charles Henderson
The Age of the Cyborg Security Team
Organizations that tried to replace humans with AI will quietly reverse course. The experiment will have run its course: productivity promises didn’t materialize, error rates went up, and attackers learned to game the systems faster than defenders did.
The wins in 2026 will come from hybrid models: humans in the loop, supercharged by AI that is tightly scoped, heavily monitored, and brutally measured. Think analysts with exoskeletons, not analysts replaced by robots.
“Human-free security operations” will be the new red flag. The smartest buyers will ask a different question:
Show me how your humans and your AI work together when everything is on fire. – Charles Henderson
Democracy Becomes a Subscription Service with U.S. Midterms
Election seasons will supercharge disinformation, deepfake campaigns, and cheap, automated harassment. At the same time, federal support and funding for election security will slide, shifting the burden to states, local governments, and private platforms that are not staffed or tooled like three-letter agencies.
The vacuum gets filled by the private sector. 2026 will see a spike in contracts for third-party monitoring of election infrastructure, dark-web chatter tracking, and rapid-response misinformation takedowns. Civic trust will have a line item, and private vendors will be the ones cashing the checks. – Cory Marchand
If You Are Not Hunting Identity Abuse, You Are Not Hunting
Attackers are done trying to break down the front door when they can simply borrow your keys to easily gain entry.
In 2026, threat hunting that does not start from identity is theater. The real fights will be in:
- OAuth and app consent abuse
- Conditional access glitches
- SaaS tokens that never expire
- Cloud control planes that log “success” while getting quietly owned
The breach reports of 2026 will read like identity horror anthologies. The teams who win will be the ones whose hunters think like fraud analysts, not just endpoint chasers. – Cory Marchand
Russia’s Hybrid Campaigns Turn Vendors into Soft Targets
You do not have to breach a government directly – just breach its vendors, researchers, and NGOs.
Russian state and state-adjacent groups will keep leaning into hybrid campaigns: with cyber, disinformation, and intimidation all woven together. Government-adjacent vendors, think-tanks, research orgs, and policy shops will get hit as a cost-effective way to influence and destabilize.
“Government vendor compromise” will become such a common phrase that it gets its own acronym and procurement checklist. – Cory Marchand
Hacktivism Becomes a Volatile Weather Pattern
The Israel-Gaza conflict will keep hacktivist energy high. Most of it will live in the DDoS, defacement, and opportunistic doxxing tier, but the spikes will track kinetic events in near real-time.
For organizations with Middle East exposure or strong public stances, uptime will start to look like a weather report. Clear skies until the next geopolitical shock, then sudden storms of nuisance-level attacks that still take brands and services offline.
Smart defenders will not “harden and hope.” They will build lightweight surge playbooks: minimal viable defenses you can activate fast when the political wind shifts. – Cory Marchand
Regulators Turn “Prove It” Into a Business Model
The SEC cyber disclosure rules are already making boards nervous. 2026 is when they start signing checks.
Four-business-day 8-K disclosures and governance transparency will push executives to ask for receipts:
- Show me that we can detect.
- Show me that we can respond.
- Show me that we tested it.
Gap analysis, validation hunts, purple team exercises, and “post-incident attestations” will stop being nice-to-haves. They will become preconditions for board sign-off. Every enforcement action that cites “insufficient controls” will quietly help sell another validation SOW. – Cory Marchand
Cyber Insurance Quietly Becomes the New Standard
Carriers are tired of paying for preventable ransomware and business interruption. Pre-bind questionnaires and post-bind validations will get stricter, more detailed, and more prescriptive.
By the end of 2026, insurer control checklists will look suspiciously like de facto standards for the SMB and mid-market. If you want coverage, you will deploy the controls. If you want better terms, you will prove they work.
For service providers, that means one thing: assessments, tabletop exercises, and validation hunts explicitly tuned to “what your carrier cares about.” – Cory Marchand
AI-Boosted Attackers Force Defenders to Prove Their Own AI Value
Attackers will happily use AI to write lures, clone voices, and scale identity attacks to a level humans cannot match. At the same time, some security leaders will use AI as an excuse to shrink teams in the name of “efficiency.”
The result is predictable: more AI-driven attacks, fewer humans to respond, and a lot of security leaders scrambling to contain the damage. Outsourcing will surge for burst capacity: triage hunts, weekend incident coverage, and “assurance as a service” that can flex up when the noise spikes.
Defenders who claim AI assists their services will have to prove it with tangible outcomes, not slideware. – Cory Marchand
The Ukraine War Makes “Partial Outage” the New Normal Test
Drone strikes, kinetic hits on infrastructure, and cyber operations will continue to blur together in Ukraine and beyond. The rest of the world will feel the blast radius in the form of power glitches, telecom degradation, and supply chain weirdness.
Continuity plans that assume clean failover and perfect backups will look naive. The real question becomes:
How does your organization operate when communications are flaky, a key supplier is offline, and you are defending while half your visibility is gone?
Assessments and pressure-test simulations that include “resilience under partial outage” scenarios will become standard for serious buyers. – Cory Marchand
Old Vulnerabilities Return Wearing New AI Masks
The industry loves to think it “solved” certain classes of problems. Then we bolt new technology on top of those problems and re-open the same wounds.
In 2026, we will see ancient issues come back with AI lipstick: unauthenticated interfaces, weak secrets management, broken auth flows, and insecure defaults – all wrapped inside shiny new platforms and agents. The stack will look modern. The vulnerabilities will make it look like 2003.
Without hard guardrails, we are going to rerun old mistakes at AI scale. – Neil “Grifter” Wyler
Code Generated Faster Than We Can Teach People to Read It
Code generation tools will keep shipping features faster than we can ship secure coding training. People with minimal experience will be writing production-impacting logic powered by AI they do not fully understand.
The dangerous part is not that the code compiles; it is that no one is truly reviewing it. The result will be a ballooning attack surface full of copy-pasted patterns and chain-of-calls vulnerabilities that no one wants to claim ownership over.
The new red team game: find the glue code no one remembers writing and assume it is full of holes. You will be right more often than not. – Neil “Grifter” Wyler
“AI Everywhere” Roadmaps Become 2026’s Most Expensive Refactor
Organizations spent 2025 shoving AI into every workflow that would hold still. In 2026, they will discover the bill.
Roadmaps that promised “AI in every process” will collide with reality: misaligned models, broken trust with users, regulatory questions, and outages rooted in AI dependencies no one can safely roll back. The fix will not be a quick patch. It will be a painful unbundling of AI from places it never belonged.
The most mature organizations will not be the ones with the most AI. They will be the ones with the least unnecessary AI and the clearest story for why the remaining deployments exist. – Neil “Grifter” Wyler
Threat Hunting: The New Red Team for Mid-Market
Not everyone can afford a full-scale red team, but everyone is tired of useless static pentest PDF reports.
In 2026, mid-market buyers will start replacing a chunk of traditional testing with recurring, hypothesis-driven hunts:
- “Assume an attacker stole a contractor’s VPN creds. Can we see it?”
- “Assume a rogue admin created a shadow SaaS app. How would we find it?”
- “Assume an AI agent is exfiltrating via allowed APIs. What would that look like?”
This will blur the line between point-in-time assessment and ongoing operations. The teams that can turn those scenarios into repeatable, defensible hunt packages will own the defensive services conversation. – Neil “Grifter” Wyler
