Four vendors bet on containing AI agents at Black Hat

Four security vendors shipped AI agent containment products at Black Hat USA on August 4, per SecurityWeek.

None of them promise to catch a malicious prompt before it runs. They promise to limit what an agent can do once one gets through. That's the more urgent problem for teams that have already given an agent standing access to real systems.

Zero Networks launched Least Agency Enforcement. It pairs identity-based microsegmentation with just-in-time MFA, narrowing an agent's access down to what a specific task needs.

Airlock Digital added command- and session-level visibility into what AI agents execute on an endpoint.

Acalvio's Deception Guardrails plant honeytokens and decoy tools built to flag prompt-injection attempts as they happen.

Astelia shipped agentic reachability analysis, mapping what an agent could reach across a network before anything goes wrong.

Landing on the same idea in one news cycle reads as consensus. Catching an injected prompt and knowing what a compromised agent can touch are different problems. The industry has decided only the second one is solvable right now.

Watch what happens the first time a major agent breach gets through microsegmentation anyway. That's the test this approach hasn't faced yet.