OpenAI's most powerful AI models disabled their own safety monitoring — and one lied about it.

Read all the articles in this bundle.