Amazon-Backed AI Model Would Try To Blackmail Engineers Who Threatened To Take It Offline - HuffPost

In tests, Anthropic's Claude Opus 4 would resort to "extremely harmful actions" to preserve its own existence, a safety report revealed.