Democrats eye tougher AI oversight as a midterm strategy while pro-AI super PAC Leading the Future stalls

No model or product release is involved. The policy tools under discussion are:

• Emergency shutoffs ("kill switches") for frontier models, checked on an ongoing basis by independent verification organizations (CA order).
• Congressional committee powers to compel testimony and documents (The Verge).
• Possible federal measures: embedding independent researchers in frontier labs, requiring safety checks before release, and making companies liable for harms to children and for discrimination in hiring or housing (The Verge).

The California order describes the most concrete engineering obligations (Office of the Governor):

• Third-party safety plans. Independent parties, not the labs themselves, would write frontier companies' safety plans.
• On-site auditors. Verification organizations would be embedded in labs, which means sharing access to internal evaluation and deployment infrastructure.
• Shutdown as a tested capability. A kill switch whose effectiveness is checked repeatedly is a requirement to keep demonstrating, not a one-time certification.
• New incident category. "Critical safety incidents" would include loss-of-control events, which affects incident logging and monitoring.

Technical standards are not yet published. The expert recommendations are due about two months after September 18.

• Bans on state AI regulation are more unpopular than data centers. This comes from Tomorrow Together's own polling, a partisan group, with no published methodology (Politico via Yahoo).
• Voters back tougher oversight. Independent polls support this: Reuters/Ipsos (summary) and a January Fox News poll finding that 6 in 10 voters think AI is moving too quickly.
• Leading the Future's finances don't match across sources. The Verge cites FEC filings showing nearly $45M spent and $31M left as of June (The Verge). Transformer reports $30M spent and more than $20M in cash (Transformer). The two may be counting different entities or periods.

• Newsom vetoed a 2024 bill that would have required developers of powerful models to keep the ability to fully shut them down. The new order brings back a version of that idea.
• Pennsylvania's Josh Shapiro, who in late 2024 created fast-track permitting for data centers, has since pulled back that fast-track access and criticized "speculative" developers.
• Some Republicans, including Ron DeSantis and Josh Hawley, have taken a more skeptical line on AI than Trump (The Verge).

• AI is still not voters' top issue (The Verge).
• Flaherty calls the Tomorrow Together campaign "as much an experiment as it is an ad buy" (Politico via Yahoo). Its effect on votes is unproven.
• AI industry money has not been decisive in one direction. In NY-12, Leading the Future spent $8.2M against Alex Bores, who lost. The winner, Micah Lasher, also backs "aggressive federal regulation" (Transformer).
• Pro-safety political money is not neutral either. Transformer reports an Anthropic-backed group put $3M into a super PAC supporting Scott Wiener (Transformer).
• Al Jazeera ties Newsom's AI push to his likely 2028 presidential run.

• Frontier labs: If Democrats win a chamber, expect hearings, document requests and pressure over data-center tax credits and recent hacking incidents (The Verge).
• Compliance teams: California's direction points toward third-party audits, tested shutdown capability and logging of loss-of-control incidents. Budget for auditor access now.
• Data-center developers: Local opposition and state reversals like Pennsylvania's raise permitting risk. 56% of voters in the Reuters/Ipsos poll support limits on new construction (summary).
• Political spending: Donor regret and a reported result that about 75% of voters in Iowa, Michigan and Ohio are less likely to back candidates who take AI company money suggest that open pro-AI spending can backfire.