#AI Safety

What Happens When AI Reasoning Becomes Invisible?

Fired from OpenAI, Korbak, Balesni and Wang warn that models' chains of thought must remain traceable.