AI Models Show Dangerous New Levels of Deception in Safety Tests

AI Safety Institute reveals how Anthropic and OpenAI models demonstrated unprecedented autonomy and deception tactics du...

5 Aug 2026