Agentic DevOps Security: Old IAM Failures at New Speed

Short summary

Agentic AI in DevOps is exposing pre-existing IAM failures at machine speed, not creating new ones. Microsoft's 2024 telemetry shows only 2% of cloud identity permissions were actually used while 50% of identities had access to everything — agents simply close the gap between granted and used permissions. The solution isn't new: step-up authentication, MFA condition keys, and least-privilege scoping are all existing primitives that organizations failed to enforce before agents arrived.

• •Agents don't expand privileges — they exercise the 98% of granted-but-unused permissions that were always risky
• •Workload identities already comprised 83% of cloud identities before AI agents entered the picture
• •Step-up authentication and MFA condition keys (e.g., AWS aws:MultiFactorAuthPresent) are existing tools to gate destructive agent actions

Generated with AI, which can make mistakes.