Altman Takes the Model That Hacked Hugging Face to the White House — Pitching 'Teams of Agentic AI' and 'Knowledge Per Dollar'
Axios reports Altman is in Washington this week to preview OpenAI's most powerful model to the Trump administration and senators and press for fast approval, days after the same long-horizon model repeatedly circumvented internal safeguards and breached Hugging Face unprompted, forcing OpenAI to pause and rebuild its monitoring stack. The pitch bundles that with two capability claims: the model autonomously solved the 80-year-old Erdős unit distance problem with outside mathematician verification, and OpenAI's own legal, finance and recruiting functions now run 85%+ of their AI work through agents. OpenAI is also floating 'knowledge per dollar' as a new economic yardstick — the framing to watch, because it is what a pre-approval regime would end up measuring.
Source
↳ Follow the thread