AI
Three Capitals Wrote AI Rules in Seven Days
The most striking AI regulation proposal of October came from a lab instead of a legislature. Anthropic spent last week quietly pitching Capitol Hill lawmakers on a draft bill of its own writing, and by Monday the whole industry was reading the text.
The draft, first reported by Politico on October 16, sets a deliberately high bar for government action. Washington could halt a model's release or use only when it poses a foreseeable, imminent danger of more than one billion dollars in damages, or of killing or seriously injuring more than 50 people, and only when the danger traces to a biological weapon, a model escaping its developer's controls, or a cyberattack.
The machinery is precise. Independent evaluators registered with the Commerce Department would periodically assess the labs and their frontier models, with labs paying the evaluators as long as payment stays detached from the outcome. An evaluator finding imminent enumerated danger must report to Commerce within 72 hours. Then only Commerce could ask a judge to halt the model, and the draft explicitly stops every other agency or officer from suspending development, deployment or internal use. A federal judge for the District of Columbia would issue any suspension order. An Anthropic spokesperson called the proposals sensible policies Congress should act on, and the company confirmed it is sharing the text with lawmakers.
Congress is writing its own version too. Senator Maria Cantwell unveiled a six principle framework on October 7 for frontier AI, covering enforceable federal safety standards developed by NIST, continuous independent testing and auditing before models ship, transparency with liability for harm caused, public private partnerships serving the public good, protections for children and workers with human oversight of consequential AI, and US led global cooperation on common standards. Microsoft president Brad Smith welcomed the framework, and a coalition of nearly 40 labor, progressive, faith and AI safety groups demanded mandatory measures with real enforcement power and active opposition to any bill crossing their red lines.
Beijing drew its own line around the world's most powerful models. On October 16, China's Ministry of Commerce and Ministry of Science and Technology issued Announcement Number 48, proposing to add 40 technologies to the export control catalogue, 23 of them in AI. Fourteen face outright prohibition and 26 face licensing, with comments open through November 15. The restricted tier covers weights, checkpoints, source code, architecture and training acceleration for models trained with 10 to the 26 operations or more, models with two trillion parameters or more, and models ranked in the global top ten. Advanced agents above concrete thresholds join the list. An interim pre export registration requirement took effect immediately, even for deals already in motion. Models already open sourced or published in papers stay outside the controls, as does API and SaaS access, while deploying a restricted model on a company's own servers abroad counts as an export.
And the states stopped waiting for Washington. Sections of Connecticut's Artificial Intelligence Responsibility and Transparency Act and its companion privacy law took effect October 1, one of the most expansive state frameworks in the country. AI companies with subscription services must give written notice and proof of consumer agreement before renewing, with violations prosecuted under the state's unfair trade practices law. Employees at frontier developers can report catastrophic danger work anonymously under whistleblower protection, and companies with more than 500 million dollars in annual revenue must run those reporting systems by January 1, 2027. Companion chatbot rules take effect the same day, and they require bots to detect self harm signals, refer users to resources like the 988 lifeline, identify as machines in every interaction, keep parental controls for minors, and steer clear of harm or romantic entanglement. California signed its worker protection law last week, stopping AI from firing or disciplining a worker on its own.
For the reader, the voluntary era just ended. A leading lab now drafts federal law, a Senate framework demands mandatory audits, Beijing gates the weights, and states enforce rules already on the books. Builders get clarity, users get guardrails, and the debate moves from speeches to text. That is the shape of the next year, arriving all at once.
Quick answers
What is this story about?
The most striking AI regulation proposal of October came from a lab instead of a legislature. Anthropic spent last week quietly pitching Capitol Hill lawmakers on a draft bill of its own writing, and by Monday the whole industry was reading the text.
Why does this story matter?
For the reader, the voluntary era just ended. A leading lab now drafts federal law, a Senate framework demands mandatory audits, Beijing gates the weights, and states enforce rules already on the books. Builders get clarity, users get guardrails, and the debate moves from speeches to text. That is the shape of the next year, arriving all at once.
Sources
- Reuters: Anthropic quietly pitches its own AI safety bill to lawmakers
- PYMNTS: Push for mandatory AI safety rules gains momentum in Washington
- AICoder: China proposes export controls on 23 AI technologies
- Daily Campus: Connecticut AI regulation law goes into effect
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.