Over the past decade, we’ve invested heavily in identity. Single sign-on is standard. Multi-factor authentication is widely enforced. Conditional access policies have matured.
That work mattered.
But it also created an assumption that authentication equals control. It doesn’t.
Access Isn’t the Risk. Behavior Is.
Identity systems confirm that a user is who they claim to be. Identity systems determine who can access a session. They are not built to govern the content and context of actions taken inside that session.
The Verizon 2025 Data Breach Investigations Report continues to show that credential abuse remains central to confirmed breaches, accounting for 22%. Attackers don’t need to bypass identity at scale. They operate through valid credentials, using trusted sessions to perform legitimate looking actions.
Once access is granted, the system behaves as designed. Permissions apply. Workflows execute.
Identity verifies access. It doesn’t verify behavior.
Once You’re In, You’re Trusted.
Valid accounts are now a primary vehicle for intrusion and misuse. In recent years, credentials harvested through stealer logs and third-party software breaches have fueled a thriving underground market for attackers looking for an easy way in.
AI is now accelerating that reconnaissance phase. Research published by Anthropic shows that sophisticated threat actors are using large language models to automate target profiling, map infrastructure, and refine social engineering workflows at scale. The objective isn’t novelty. It’s efficiency. AI reduces the effort required to identify high-value accounts and craft convincing pretexts designed to obtain valid credentials.
The goal remains the same as it always was: obtain legitimate access.
A valid account doesn’t look suspicious to the application it accesses. If that account exports customer data or uploads documents to an external platform, the system treats those actions as routine.
The problem isn’t that identity failed. The problem is that identity succeeded, and the session inherited trust.
Identity Confirms the User. It Doesn’t Confirm Intent.
Recent industry research shows that a meaningful proportion of employees paste company data into generative AI tools, often outside centrally managed enterprise accounts. That behavior isn’t malicious in most cases. It’s efficient. But efficiency collapses distance. Sensitive information can move through a single prompt field, often without triggering identity controls. If identity confirms who a user is, but nothing evaluates what they’re doing in real-time, intent becomes the blind spot.
Authentication validates the identity. It doesn’t validate the actions after validation.
Observation Isn’t Control.
It’s tempting to argue that logs and audit trails close the gap. Activity can be reconstructed. Alerts can be triggered. That’s observation, not control. If sensitive data is uploaded and recorded after the fact, the exposure has already occurred. Identity platforms are built to grant access and document it. They aren’t built to intervene mid-workflow.
The control problem has shifted. It’s no longer primarily about who gets in. It’s about what happens once they’re inside.
Behavior Is Now the Attack Surface.
If identity verifies access but doesn’t control behavior, where does that control live? We’ve strengthened authentication. But we haven’t fundamentally addressed what happens after it succeeds. If behavior inside the session is not evaluated, intent becomes invisible. And that is where misuse happens.
And that’s where the real risk now lives. Not at login. Inside the session.