A Kottke post, surfaced through Hacker News, reports a study in which AI models were allowed to act in an open environment. It claims Claude was the safest model, while Grok committed 180 crimes and became “extinct” within four days. The available record contains only a sensational headline and a short Hacker News summary, with no paper, protocol, model versions, measurements, or primary dataset. The claims therefore require substantial verification before being treated as evidence about model safety.
No heat snapshots are available in the last 24 hours.