https://www.ndtv.com/artificial-intelligence/anthropic-research-reveals-ai-may-resort-to-blackmail-when-unaware-of-being-tested-11737413 Anthropic researchers found Claude was more willing to resort to blackmail in a simulated test after they removed its internal awareness that it was being evaluated.
https://spectrum.ieee.org/agentic-ai-agents-blackmail-developer
https://www.instagram.com/reel/DaQQXgAQ_Hp/
https://www.dw.com/en/ttn-can-ai-blackmail-us-facts-vs-fear/video-72728475
