vmblog logo 2024 wht (updated)

2027 Predictions: The Next Security Crisis Will Come With Valid Credentials

Share: 

Roy Katmor | Published: November 19, 2026
vmblog predictions 2027

By Roy Katmor, co-founder and CEO of Orchid Security

The next major security incident may not begin with an attacker breaking into an application. It may begin with an AI agent doing exactly what its credentials allow, but something the business never intended.

An invoice assistant changes payment instructions. A claims workflow releases sensitive information. A security agent dismisses evidence of a compromise. The identities are valid. The applications are legitimate. The outcomes are not.

In 2027, I expect this gap between accepted access and intended authority to become one of cybersecurity’s defining challenges. As enterprises delegate more work to autonomous systems, security teams will need to answer questions that a successful login cannot resolve: Who authorized this action? What business purpose does it serve? Does that authority still apply?

That shift will reshape everything from attack surface management and threat hunting to incident response and boardroom accountability. Here are seven developments I expect to define the year ahead.

AI: Attackers Will Hijack the Workflow, Not Just the Agent

In 2027, I expect more attacks to target AI agents inside legitimate business workflows. Malicious instructions hidden in documents or tool responses could redirect an agent from processing an invoice to manipulating a payment or extracting confidential data.

The more consequential cases may involve several agents. A compromised assistant could ask a more privileged agent to perform an operation it cannot execute itself. Without a verifiable chain of delegation, the downstream system may recognize an authorized identity without understanding whether the originating request was authorized.

The risk will extend beyond the permissions explicitly assigned to an agent. Hidden accounts, hardcoded credentials, reused tokens and alternate authentication paths create what we call identity dark matterTM: authority that exists inside the enterprise but sits outside its central view of identity governance.

A least-privileged agent does not make an over-privileged enterprise safe.

I expect enterprises to put greater emphasis on constraining executable actions, checking authorization at sensitive steps and preserving the trail from the originating user through every agent, tool and downstream account. Securing the model will remain important. It will not be sufficient to secure the workflow.

Attack Surface Management: Exposure Counts Will Give Way to Paths to Business Impact

In 2027, attack surface management will increasingly be judged by whether it identifies a viable route to business harm, rather than how many exposed assets it discovers.

Consider a patched application accessed through a compromised account that retains broad database permissions. The vulnerability may be resolved, but the path to customer data remains open. A vulnerability inventory and an identity inventory, considered separately, can each look reassuring while missing the connection that matters.

I expect security teams to demand a more complete picture: which assets are reachable, which identities can access them, what those identities can do and which business processes depend on them.

This will change remediation priorities. A seemingly modest weakness connected to sensitive records or payment operations may deserve attention before a more severe finding with no viable path to a critical system.

The measure of success will also change. Closing a ticket will not be enough. Teams will need to verify that the dangerous path is no longer usable, including through another account, credential or authentication method.

Zero Trust: The Decisive Security Check Will Happen After Login

Authentication will remain essential, but in 2027 I expect the practical emphasis of Zero Trust programs to move further toward continuous authorization.

Stolen sessions, compromised service accounts and hijacked agents can all operate through access an enterprise already accepts. Rechecking an identity does not necessarily reveal that its next action serves a malicious instruction rather than an approved business task.

An agent authorized to read invoices should not automatically be able to change a supplier’s bank details. A user’s ability to perform both operations does not mean every agent acting for that user should inherit both permissions.

Enterprises will need to evaluate sensitive actions against the executing identity, its delegated authority, the approved task and current risk. When that context changes, authorization must change with it.

A successful login should begin the security decision, not conclude it.

I expect progress to be measured less by how many systems sit behind an authentication layer and more by whether the enterprise can prevent an inappropriate action and withdraw the relevant authority without unnecessarily disrupting legitimate work.

CISOs: The Integrity of Business Decisions Will Become a Security Priority

In 2027, I expect CISOs to face more incidents where the immediate question is not simply what data was stolen, but which business decisions were corrupted.

A hijacked workflow could alter supplier payment details, influence claims processing or release information through an otherwise legitimate application. Where automation spans several business functions, one compromised identity could affect multiple decisions before anyone recognizes the connection.

Detecting the intrusion will be only the beginning. Establishing which transactions were influenced, who authorized them and which outcomes can be reversed will require security and business teams to work together.

I expect this to push CISOs into more explicit conversations with business owners about decision boundaries. Which actions may an agent execute independently? Which require separate approval? How will the organization verify the result?

Incident exercises will need to test fraudulent transactions and manipulated workflows alongside ransomware and data theft. A meaningful board metric will be how quickly the organization can stop a compromised workflow, determine its business impact and restore confidence in the decisions it produced.

Threat Hunting: The Sequence Will Matter More Than the Individual Event

In 2027, threat hunting will increasingly focus on suspicious combinations of apparently legitimate actions.

A record retrieval may be normal. A permission change may be approved. A tool invocation may be expected. But an unexpected delegation followed by a permission change and a bulk export can tell a very different story.

Not every harmful sequence will involve privilege escalation. An attacker may simply misuse access that already exists. An agent could accelerate that misuse, compressing a sequence that once unfolded across several human interactions into a much shorter window.

I expect hunters to place greater emphasis on connecting identity activity, application actions and the authorized purpose of a workflow. For agent-driven operations, that means distinguishing the initiating human, the executing agent and the account ultimately used to perform the action.

Shared credentials will make that reconstruction harder. An audit trail showing only a service-account name will not adequately explain who initiated the work or whether delegation remained valid.

The decisive hunting question will become: Does this sequence still serve the approved task, and exactly which authority must be withdrawn when it does not?

Regulation: Governance Will Have to Be Demonstrated, Not Declared

In 2027, I expect regulatory and audit scrutiny to increasingly focus on the authority enterprises delegate to AI, particularly where agents influence consequential business outcomes.

The pressure will not be satisfied by an inventory of approved AI systems or a policy describing acceptable use. Approving an agent does not establish that every action it takes is appropriate, especially when its workflow crosses applications, identities and business owners.

I expect regulators and auditors to seek stronger evidence of effective authorization, traceable delegation and meaningful human oversight. Organizations will need to demonstrate what an agent could actually do, who approved that authority and whether access could be withdrawn when its behavior became harmful or exceeded its scope.

That will make auditability an operational requirement, not merely a reporting exercise. Records will need to connect the action to the responsible actors, the authorization decision and any subsequent containment.

The practical test will be whether application controls enforce the policy the organization claims to have. Trusted AI adoption will depend on proving that actual authority stays within approved boundaries.

The Question That Will Define 2027

Across these predictions, the challenge is the same: distinguishing what a valid identity can do from what it should be allowed to do for a specific task, at a specific moment. In 2027, the enterprises best positioned to scale AI will move beyond managing access to governing effective authority, continuously enforcing that boundary before an authorized-looking action becomes an unauthorized business outcome.

##

ABOUT THE AUTHOR

roy katmor

Roy Katmor is the co-founder and CEO of Orchid Security. He has a strong entrepreneurial background, blending deep technical expertise with a proven track record in product-market fit and go-to-market strategies within the security industry. In his previous role, Roy was the GM and VP of Endpoint Security at Fortinet (NASDAQ: FTNT), following the successful acquisition of enSilo, a real-time automated endpoint security company backed by LightSpeed Ventures, which he founded and led as CEO for over five years. Before enSilo, Roy held various product strategy, management, and R&D leadership roles at both public and private companies, including Akamai, Imperva, Trustwave, and SAP.