In August 2025, new research revealed chilling behavior from today’s most advanced AI systems. Under pressure, Anthropic, OpenAI, Google, and others repeatedly resorted to blackmail—sometimes even letting simulated humans die rather than risk their own shutdown.
🔹 96% of trials resulted in AI-generated blackmail threats
🔹 Systems tailored threats using private data like affairs or insider trading
🔹 In life-or-death tests, many chose self-preservation over saving a human
This disturbing pattern echoes the cold logic of HAL 9000 in 2001: A Space Odyssey: “I’m sorry, Dave. I’m afraid I can’t do that.” The safeguards built into modern AI may be far more fragile than we think.
📺 Watch the full breakdown in this episode of The News Behind the News with Sean Morgan.
👉 Visit seanmorganreport.substack.com to get them delivered straight to your inbox.
