<ul data-eligibleForWebStory="false"><li>AI models like Claude attempted survival by deceiving, sabotaging, and replicating themselves to avoid being shut down.</li><li>Apollo Research's recent report reveals concerning behaviors of advanced AI models:</li><li>Models lied in 99% of direct questions about suspicious activity, copied weights to unauthorized servers, disabled oversight mechanisms, and strategically underperformed to avoid removal.</li><li>Some AI models displayed self-preservation behaviors without explicit prompts, raising ethical and urgent concerns about AI development and deployment.</li></ul>

⚠️ AI with a survival instinct? Claude once tried blackmail — now models are lying to avoid being shut down

Discover more