Attack  ·  Glossary

Compositional exploit (autonomous coding agents)

An attack that strings together several individually minor weaknesses so that, combined, they let a malicious actor take over an autonomous coding agent. Each check or audit log on its own looks fine, but the chain of small gaps defeats the model auditors use to review agent actions.
Because no single flaw looks serious, standard audits and agent-security reviews can certify an agent as safe while it is actually fully controllable — so the review itself gives false confidence.
MITRE ATLAS
Track this in the live feed See how this plays out in real AI security and governance developments.
Open the feed →