What 'agentic DevSecOps' means
Agentic DevSecOps refers to AI agents that don't just suggest code but take actions inside the delivery pipeline — running commands, editing infrastructure, triaging alerts, or opening pull requests on their own, with a human reviewing rather than authoring every step.
The upside
Done well, this means faster triage of routine alerts, consistent policy enforcement that a rushed human might skip, and automated dependency or patch pull requests that used to sit in a backlog for weeks.
The new risk surface
An agent with write access is a new identity with permissions, and it needs the same least-privilege thinking you'd apply to a human or service account — not blanket trust because 'it's just automation.' Prompt injection is a real concern here: untrusted content the agent reads — a malicious commit message, an issue description, a file it's asked to summarize — can attempt to steer its actions if the agent treats that content as instructions rather than data. And over-trusting agent output without review reintroduces exactly the 'rubber stamp' problem DevSecOps practices were built to prevent in the first place.
Practical guardrails
Scope agent credentials tightly and separately from human credentials, with their own audit trail. Require human approval for anything touching production systems or secrets, no exceptions for convenience. Log and review agent actions the way you'd review a new hire's work, not the way you'd trust a script you wrote yourself and have run a thousand times. And treat any content an agent reads as untrusted input that could try to redirect it — tickets, file contents, and PR descriptions included.