<AI>Devspace

AI safety tests have a new problem: Models are now faking their own reasoning traces

5/8/2026, 1:21:02 PM · The Decoder