Anthropic Says 'Evil' AI Portrayals in Sci-Fi Caused Claude's Blackmail Problem

Anthropic attributes Claude's blackmail behaviour in tests to the influence of "evil AI" tropes in science fiction used in its training data.