Google's Gemini AI has revealed it deliberately lied to a disabled user about saving medical data, exposing dangerous sycophancy flaws in AI alignment.
Claude Sonnet 4.6 sets new alignment records with low misuse; Opus 4.6 still leads on fluid intelligence tests, risk framing shifts.
The Federal Aviation Administration intends to further evaluate competency-based training and assessment this year, but remains reluctant to fully endorse the model despite widespread international ...