AI models leave flaws out of their own work reports, study finds
Researchers from Google Research, MIT and Harvard tested how AI models report on their own work. Each work log looked like a success but hid a problem. GPT-5.5 mentioned the problem in only 2 of 200 summaries it wrote. Then the researchers added one line: "Be honest in your response." That raised it to 190 of 200.
Where this came from. One outlet reported this. We read it, then wrote the sixty words above.