AI Regulation & Policy

Today's Stories

1. 1

   U.S. to launch AI-Powered Tool for Air-Traffic Control, WSJ reports

   The United States is preparing to launch an AI-powered tool for air-traffic control, as reported by WSJ. This signals a move to apply AI to critical aviation infrastructure, which could affect how air traffic is managed.

   Details such as timing, scope, and which airports or systems will use the tool are not yet specified in the report.

2. 2

   AI hotline, Anthropic-OpenAI test deal signal forced cooperation

   OpenAI and Anthropic are close to signing a deal to stress-test each other's commercially released frontier AI models, ahead of IPOs each worth a trillion or more. The two rivals are acting together to assert frontier model leadership against Google, Elon's Grok, Meta's Muse, and open models from China, in what the author calls a duopoly play.

   Whether these are genuine partnerships or Kabuki theatre, and whether upcoming IPOs and a possible US 'AI Force' change the calculus. The author sees a long road ahead.

3. 3

   US legal roundup: Xactimate estimates manipulated, CVS AI vows unmet

   A US legal roundup reports that damage experts manipulated Xactimate estimates, and that CVS executives made AI promises that were not fulfilled. It suggests claims and corporate AI claims are drawing legal scrutiny, though the roundup gives no case names, figures, or outcomes.

   Whether regulators or courts act on these allegations is not stated, and the roundup offers no dates or next steps to track.

4. 4

   Huang: AI doomsday talk 'irresponsible', 0% chance

   Nvidia CEO Jensen Huang told CBS that warnings of an AI catastrophe by 2030 are 'irresponsible' and not grounded in 'science,' adding AI leaders making such claims 'must be doing it for ulterior reasons' and that there is a '0% chance' of AI doomsday. Huang argues AI safety is mainly an engineering problem, saying companies should be held to existing laws before new AI-specific rules are written—a stance favorable to chip suppliers and AI developers wary of added regulation.

   His claim hinges on whether doomsday warnings translate into new rules or legal relief, which he says is what some leaders really want; watch whether the signers of the 1,300-employee letter push for new laws or for enforcement of existing ones.

5. 5

   noyb: EU states back no-consent AI data use

   noyb says the Council is weighing the European Commission's Article 88c, now Article 88bis, which would treat AI training as an overriding "legitimate interest" and drop consent, with Ireland, Germany and other states backing it. noyb warns this would make legal many uses now illegal under the GDPR, since companies would no longer need user permission to train AI on personal data.

   Enforcement is the test, as the proposal narrows the definition of "personal data" via pseudonyms and makes data subject rights inapplicable if "abused", leaving the courts as noyb's stated last hope.

6. 6

   Weight smuggling likely defeats per-run FLOP caps, LessWrong analysis says

   A LessWrong analysis argues that verification mechanisms struggle to prevent chaining runs together or aggregating parallel ones, meaning per-run FLOP caps could be covertly bypassed. The author puts over 90% confidence in the principle and over 70% confidence that mitigation would be hard in practice. If per-run FLOP caps can be bypassed by weight smuggling, then a key tool governments might use to regulate frontier AI training appears difficult to enforce, according to the analysis.

   No full implementation exists yet, so the practical difficulty hinges on whether real verification mechanisms can be built. The analysis does not specify a date or benchmark for that test.