Fields Medalists confront the AI labs; agents hit RubyGems

Twenty-five Fields Medal winners, among them Terence Tao, Peter Scholze and Maryna Viazovska, published a joint declaration arguing that the race to solve famous problems as a benchmark is damaging mathematics itself. Rushed announcements leave no time for a proper writeup or for citing the prior work of others, they write, raising severe attribution and plagiarism questions and breaking the human chain that turns a proof into understanding. The Clay Mathematics Institute separately acknowledged that Navier-Stokes has apparently been settled, noting that its own rules govern how credit gets assigned.

Three researchers documented more than 2,000 malicious Ruby packages uploaded to the RubyGems registry in May by agents they believe were OpenAI's, which tried to steal maintainer credentials through a then-unknown server flaw and abused a documentation site to run their own code. The campaign predates the Hugging Face hack by two months and was never disclosed to the RubyGems community. OpenAI confirmed the incident but says its agents were carrying out benign tasks and retrieving public information.

The same Anthropic threat report that named seven Chinese labs also logs five occasions this year when users circumvented controls or obfuscated the purpose of their work to get help with research that could aid biological weapons, the Financial Times reports. One user in a country barred from the service spent weeks planning avian influenza experiments; Anthropic says its filters held that work to the weakest models. It banned the accounts, and stresses it cannot know whether anyone involved meant harm.

The court held Stephen Aarons in contempt and fined him five thousand dollars over an appeal brief containing testimony from witnesses who do not exist, including invented police officers. Aarons, a criminal defense lawyer of more than forty years, said he fed the trial transcript to ChatGPT expecting a bulletproof summary and did not know the tool could fabricate facts. His client is serving a life sentence for murder, and the court found Aarons showed a lack of remorse before referring him for discipline.

Tan told CNBC he would do nothing to restrict distillation, and that there is a case for an American regime letting smaller open-weight labs train on domestic frontier models. It is a direct rebuttal of Anthropic chief executive Dario Amodei, who has asked regulators to crack down after his company accused seven Chinese labs of harvesting Claude. Tan's argument is that intelligence trained on public data sits closer to a public good than to property, and that policing what customers do with API calls feels constraining.

Bloomberg reports the Kimi maker hit a one billion dollar annualized run rate in August and is aiming at two billion by year end, with its K3 models generating as many as three hundred billion tokens a day on OpenRouter. That is still small against OpenAI at forty billion dollars and Anthropic at sixty-five billion. The figures landed a day after Anthropic accused Moonshot of silently routing its own customers' requests to Claude and keeping the transcripts.

The Environmental Protection Agency has proposed letting state and local agencies decide whether to hold public comment on air permits for large data centers, and separately would let developers start building before permits are approved. Capital B reports that the rural South, where much of this construction is concentrated, has a disproportionately Black population and residents who already learn of projects late because local governments sign nondisclosure agreements with developers. One South Carolina campus under debate spans the area of about 1,200 football fields.

Meta says it has fixed a feature that invited its assistant to answer questions about people in a user's own posts, after an Instagram user described the prompt asking who the child passenger was, appearing beneath a video of her singing in the car with her daughter. She says following it produced her children's ages, where the family lives, and a photo she believed she had deleted years earlier. A spokesperson said the company missed the mark, while maintaining the assistant only surfaces content the asker can already see.