Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilities

Open in new window