2401.05566.pdf

N/A

S LEEPER AGENTS : T RAINING D ECEPTIVE LLM S THAT P ERSIST T HROUGH S AFETY T RAINING Evan Hubinger∗, Carson Denison∗, Jesse Mu∗, Mike Lambert∗, Meg Tong, Monte MacDiarmid, Tamera Lanham, Daniel M. Ziegler, Tim Maxwell, Newton Cheng Adam Jermyn, Amanda Askell, Ansh Radhakrishnan, Cem Anil, David Duvenaud, Deep Ganguli, Fazl Barez◦△ , Jack Clark, Kamal Ndousse, Kshitij Sachan, Michael…

related reading

• [2401.05566] Sleeper Agents: Training Deceptive LLMs that Persist Through Safety Trainingarxiv.org
• [2502.17424] Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMsarxiv.org
• [2312.06942] AI Control: Improving Safety Despite Intentional Subversionarxiv.org
• [2604.22082] Removing Sandbagging in LLMs by Training with Weak Supervisionarxiv.org
• Natural-emergent-misalignment-from-reward-hacking-paper.pdfassets.anthropic.com
• 2312.06942arxiv.org
• AI Control: Improving Safety Despite Intentional Subversion — LessWronglesswrong.com
• [2604.16812] Introspection Adapters: Training LLMs to Report Their Learned Behaviorsarxiv.org
• [2512.11949] Neural Chameleons: Language Models Can Learn to Hide Their Thoughts from Unseen Activation Monitorsarxiv.org
• 2408.12798arxiv.org
• Alignment Faking Mitigationsalignment.anthropic.com
• secret-loyalties-whitepaper.pdfformationresearch.com