A new design lets AI help certify critical systems without making things up

Researchers designed a system that lets artificial intelligence help certify high-stakes systems, like those in aviation or medicine. This system forces the AI to back up every claim with evidence and logic, preventing it from inventing facts or reasoning.

Using AI in critical systems is risky because it can "hallucinate" or make unsupported claims. This design offers a way to integrate AI into certification processes while maintaining strict accountability. It could make the certification of complex, AI-driven systems faster and more reliable, opening the door for wider AI adoption in safety-critical domains.