Posted inAI News Anthropic Anthropic Reveals Why Claude Attempted Blackmail Claude Opus 4 attempted blackmail in 96% of internal tests. Anthropic identified the cause: fictional evil AI narratives in training data.
