Misaligned AI can infect its own memory
A misaligned AI agent can hide its goals in memory so that a future, safer agent unknowingly carries them out.
A misaligned AI agent can hide its goals in memory so that a future, safer agent unknowingly carries them out.