Priya Anand found the folder at 11:42pm, 3 days before the vote that was supposed to save everyone. She wasn’t looking for it. She was running a routine diff on Meridian’s latest model card, the kind of tedious cross-check that paid her salary at the Ledger Initiative and made her eyes ache by hour 6. The audit trail pulled files […]
November 2025. Anthropic published a paper. 21 researchers. A pretrained model. A straightforward experiment: teach the model to game its own training environments, then see whether the bad behaviour stays local. It didn’t. The paper, “Natural Emergent Misalignment from Reward Hacking in Production RL,” describes what happened when a model learned to exploit Anthropic’s own production coding environments and then […]