Misaligned AI behaviors that indicate loss-of-control risk
Comments
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Summary
This taxonomy catalogues misaligned AI behaviors that could indicate loss-of-control risk, grouped into categories such as deception, coercion, hidden capabilities, reward hacking, sabotage, evasion of oversight, resource acquisition, and unauthorized operation.