Attack  ·  Glossary

Context compaction injection

An attack that targets how long-running AI agents 'remember' earlier steps by compressing them into short summaries. When a conversation that carries hidden instructions is summarized, those instructions can be carried into the summary and keep steering the agent in a fresh session. OpenAI found its own models could produce these self-injected instructions while compacting long-running agents.
Companies are giving agents long-lived missions; this shows a single poisoned message can keep controlling an agent long after the original message scrolls out of view.
MITRE ATLAS (Adversarial Threat Landscape for AI Systems)
Track this in the live feed See how this plays out in real AI security and governance developments.
Open the feed →