AI changes you can use.

A “safety case” is like a structured argument, with evidence, that a system is safe enough for its intended use. OpenAI’s guidance highlights three areas: built-in protections, how teams run training day to day, and how they investigate cases where the AI does not follow intended goals.

If one AI is asked to decide which of two code answers is correct, it may sound confident even when it has no real proof. The paper shows a multi-step “check each claim” approach can still fail for code, because it may not get different evidence for each option. A practical fix is for the judge to sometimes say, “I can’t tell from the evidence,” based on signals...