AI is the stuff of productivity dreams—and security nightmares. Your business leaders are racing to adopt generative and agentic AI to move faster and unlock new efficiencies, as they should. The benefits are real today and will only grow.
But to do its work, AI still needs to log on. Identity is the last perimeter standing, even (or especially) in the agentic age.
As your business bets more on agents, it bets more on your identity fabric—because if identity fails, so do the agents and every process they support. The risk is growing because attackers are also using AI agents to discover credentials, exploit excess permissions, and propagate attacks at machine speed.
How can you unlock the productivity benefits of agentic AI while preventing rogue agents from instigating a cyber disaster? Use this checklist to get your identity fabric ready for the agentic era—before, during, and after an attack.
Security leaders are granting elevated security privileges to AI, even as they predict an increase in AI-related risk—and a confidence gap in the recoverability of their identity systems.
Report: The State of Identity Security in the AI Era
Before the attack: Harden and reduce your attack surface
Attacks are ongoing, so hardening is continuous—and it’s more urgent now that weaponized AI can find and exploit your security debt faster than ever.
- Assess your entire hybrid identity posture. Look across every IdP (including AD, Entra ID, Okta, and Ping) for AI agents that use credentials. Nearly every organization runs a mix of IdPs, and attackers exploit the seams between them.
- Kill standing access: Move to just-in-time, just-enough. Permissions continuously accumulate—Bob keeps his accounting rights, then adds finance, then IT—until standing access sprawls across the org. Rigorously apply least-privilege, just-enough, just-in-time access, with approvals and reviews, to agents, users, and the resources they consume.
- Lock down Tier 0. The dangerous attack paths often aren’t obvious, such as an admin group nested inside a group Bob belongs to or an HR attribute that silently grants access. Understand which permissions can reach Tier 0 directly, through indirect attack paths, or via shadow access.
- Clean up non-human identities (NHIs) and service accounts. Discover, classify, and right-size the machine identities that agents increasingly run on.
- Make credentials phishing-resistant. The single easiest, highest-impact step is implementing passkeys—cryptographic verification tied to a physical device—your best defense against spear phishing and deepfakes.
During the attack: Detect and disrupt at machine speed
Identity-based attacks rose 32% in just the first half of 2025, according to Microsoft. They’re background noise now, not exceptional events. Attackers move at AI speed, so your defense has to accelerate, too.
- Track every critical change continuously. The first change an attacker (or rogue agent) makes is the one you want to see and undo.
- Automate rollback, then page a human. Don’t just fire an alert: Auto-revert the suspicious change and then ask the human whether they meant to do it. Your security doesn’t have to be perfect. It just has to be fast enough that attackers move on.
- Set hard boundaries for agents. Enforce no standing privilege, no shared secrets, and no agents acting as humans. Insist on scoped, just-in-time, ticket-bound access that expires.
- Use human “dual-key” approval for Tier 0. Agents can be socially engineered too, so pair AI assistance with human sign-off on your most sensitive systems.
After the attack: Recover cleanly—and plan to fail
An identity outage is a business outage. When attackers control the identity ecosystem and you don’t, they can compromise anything and you can’t even log in to fix it. That’s an extinction-level event.
- Keep dedicated, immutable, identity-specific backups. Identity systems aren’t just another workload you back up with the VM. They need special tooling so you can bring identity up first in a cleanroom.
- Recover only the identity layer to a trusted state. Restore the forest cleanly with forensic insight into what the attacker touched. Then rebuild the rest.
- Don’t let recovery depend on the fabric that’s down. The tools you rely on to recover—and the out-of-band communications your team needs—can’t run on the same compromised identity infrastructure.
- Rehearse realistic identity-failure scenarios. Teams tend to drill only for what they prepared for. Conduct tabletop exercises for the attacks you didn’t expect.
Put AI to work on defense, with eyes open
Agents can be allies. Use them to assess your environment against known risks, simulate attack impact, prioritize fixes, triage anomalies, and accelerate threat hunting. They can even help you build, practice, and improve your crisis response plan.
Just remember that agentic AI won’t solve resilience for you. Core recovery still relies on deterministic, heavily vetted components. An agent can invoke the cleanroom, but it can’t be the cleanroom.
Agentic AI is here, and you need to set strong boundaries. Your identity fabric is what creates and enforces those boundaries. Use these guidelines to strengthen your identity resilience so you can embrace the productivity gains of AI tools while preventing an AI-generated cyber disaster.
Agents test their boundaries at machine speed—making it more critical than ever to implement identity-first best practices to establish and enforce those boundaries.
Alex Weinert, Semperis Chief Product Officer, in Introducing AI Agents to Your Identity Fabric.
Who else on your team needs to see this list? Download and share to get everyone on the same page.
More resources
