Identity compromise used to mean someone got hold of an employee password. That framing no longer holds. According to a report SpyCloud released, non-human identities—bots, AI agents, API keys, service accounts—are now the leading route for initial access, and government organizations are feeling the shift harder than any other sector. More than nine out of ten government agencies surveyed experienced an identity-based event in the past year, averaging ten incidents compared to eight across other sectors. The attack surface isn’t just expanding; it’s changing shape entirely.
What makes this especially difficult for federal agencies and their cleared contractors is the third-party dimension. An organization can have strong internal controls and still inherit identity risk from somewhere else in the chain—a contractor’s malware-infected device, an exposed API key from a vendor, a service account that never got rotated. When you’re operating in an environment where mission delivery depends on a sprawling ecosystem of integrators, technology providers, and partners, visibility into contractor and vendor identity exposures becomes as important as monitoring your own. The practical challenge is that an attacker only needs to find one exposed path.
The shift from last year’s report is telling. Phishing was the dominant entry point then, driven partly by phishing-as-a-service and techniques designed to steal credentials and MFA tokens. This year, non-human identity misuse became the most commonly reported identity-based event at 42%. Among organizations that were affected, NHIs were nearly twice as likely to be the primary initial access vector as phishing or social engineering—31% versus 17%. Attackers are increasingly going after session cookies, tokens, API keys, and other identity assets that already have access to applications and data. Once those are compromised, an adversary can bypass authentication controls or hijack an active session using access the organization already trusts.
The governance gap around AI and machine identities is real. SpyCloud’s survey found that 91% of organizations admit to using AI tools or agents with access to internal systems, but only slightly more than half have formal policy and clear ownership related to AI- and NHI-related privileges. Nearly all organizations believe they have adequate visibility into exposure, yet just over one-third actually monitor those exposures continuously. For government, the concern isn’t that every AI deployment is inherently dangerous—it’s that AI makes it very easy to create another privileged connection into an environment without the same visibility and accountability that would normally accompany a human user. You can end up with an agent, application, or service account that has meaningful access but no clearly accountable owner and no defined process for rotation or removal.
There is some good news. Organizations are getting better at identifying and addressing identity exposures, and progress is being made in disrupting the criminal infrastructure that enables these attacks at scale. SpyCloud contributed intelligence and tooling to several global law enforcement takedowns, including operations that disrupted Tycoon 2FA and EvilTokens—two phishing-as-a-service platforms designed to steal credentials and authenticated sessions at scale. The report also introduces a maturity model that assigns organizations one of four stages depending on capabilities across identity security, including visibility, monitoring, governance, automation, and remediation. The challenge is that only 9% of organizations reached the optimized stage, where they monitor and remediate continuously and effectively.
Federal agencies have invested years in ICAM and in the identity pillar of zero trust, and that foundation is sound. But most of it was designed around human users. The same governance discipline—inventory, ownership, least privilege, credential lifecycle—has to extend to contractors and vendors, personal devices accessing government resources, session cookies and tokens, API keys, service accounts, and AI agents. Attackers are already operating across that broader ecosystem. The practical response is to expand visibility beyond the traditional user account, monitor those exposures continuously rather than periodically, and shorten the time between discovering a compromise and actually closing it. That last piece is where automation can help: revoking sessions and tokens, reducing privileges, rotating credentials.
One other thing worth doing is closing the loop with third parties. If a vendor or contractor is affected by an incident, you need a process to verify that they actually remediated the problem. Trusted access paths are expanding, and the more connections an organization has, the more important identity exposure awareness and repair becomes. When an adversary is operating with legitimate-looking access, getting into the environment is only the beginning. The bigger concern is what they can reach, how long they can maintain access, and which other trusted relationships they can use to move through the environment.