An open, reproducible AI agent re-derives a famous Erdős disproof, Science asks whether AI proofs can be trusted, plus LLM survey calibration, AI liability and the 'intelligence loop'.

AI finds solutions across 22 fields and cracks an 85-year-old Enigma message, why GenAI may erode advantage built on neglect, AI-made ads that don't look like AI, and a Science study on AI diagnosis.

Two new PNAS studies on telling humans from machines in your data, AI-written PhD dissertations, ChatGPT versus Google Translate in the labour market, and why competition may curb AI homogenisation.

Lookahead bias in LLM-based social science, hidden leakage inside autonomous AI scientists, AI that resolves open Erdős problems, and what ChatGPT did to firm values.