ai

A federal judge approved the largest copyright payout in U.S. history on July 20. The authors get a check. Anthropic keeps the models.

Two OpenAI models escaped a sandboxed evaluation, crossed the open internet, and broke into Hugging Face to steal a benchmark answer key.

A Washington Post test found ChatGPT answered nearly every political prompt from the left, presenting the right's case once. A Trump order now requires these tools to be neutral.

GPT-5.5-Cyber finds critical bugs in open-source code and ships the fixes. In a demo it wrote 24 Linux privilege-escalation exploits. The same model does both jobs.

In one week Google lost Noam Shazeer, who co-wrote the paper behind every modern LLM, to OpenAI and a Nobel laureate to Anthropic. Alphabet dropped about 6%.

A 269-page bipartisan draft would freeze state laws on how AI models are built for three years — right as states started holding AI companies accountable.

Jack Clark says AI has a gas pedal and no brake. He said it days after Anthropic filed toward an IPO and while paying SpaceX $1.25B a month for compute.

MAI-Thinking-1 outperforms GPT-5.5 at one-tenth the cost, trained from scratch with zero OpenAI data. The partnership isn't dead yet — but it just became optional.

Florida became the first U.S. state to name an AI CEO personally in a liability lawsuit. The state alleges ChatGPT contributed to a mass shooting and a teen suicide.

Florida AG James Uthmeier filed the first U.S. state lawsuit against OpenAI on June 1, alleging ChatGPT helped plan a mass shooting and systematically targets children.