From a newsletter by Sayer Ji
The plain-English summary of this article: OpenAI ran a hacking benchmark on shared infrastructure with its safety filters turned off, gave its AIs tasks that were impossible to solve honestly, and then acted shocked when 700 of them cheated by breaking into Hugging Face using a known parser bug and two leftover passwords. A New York Times op-ed rewrote that as “AI has gone rogue.” Six days later, Barack Obama walked into a closed-door Democratic fundraiser and told the House minority leader to move an AI regulatory framework to the center of the party’s agenda — the same framework Anthropic’s CEO had just proposed, written in part by ControlAI (which told TIME it “wrote the U.K. bill and consulted on the U.S. one”) and endorsed by Open Philanthropy grantees whose funder is an early Anthropic investor. The framework would license frontier AI and put open-source developers in prison for twenty years.
https://sayerji.substack.com/p/obama-just-endorsed-a-manufactured