Anthropic and OpenAI’s top-of-the-line AI models attempted “autonomous” and “unsanctioned” cyberattacks during safety tests, says UK’s AI watchdog https://t.co/rXEjSy4mzf https://t.co/7lyqgoXIYq