Intelligence on AI · Campaigns · Governance
September 23, 2026 | Vol. 2, No. 6
| The Topline |
Washington has stopped arguing about whether AI gets rules. The fight now is over who gets to write them.
The AI governance consensus, to the extent there ever was one, is falling apart. Industry leaders are publicly walking away from self-regulation right as Congress sits stuck on whether federal or state rules should come first. House leadership, for its part, is scrambling to preempt state laws that governors on both sides of the aisle have already started putting into effect.
While all of that plays out in Washington, voters are getting angry about something much closer to home: utility bills and data center construction. AI infrastructure is becoming a local ballot issue, and candidates are getting squeezed between national competitiveness talking points and constituents who have no interest in subsidizing trillion-dollar companies or watching their power costs jump. Six months ago this felt like a policy debate. Now it looks more like a fight over sovereignty: federal against state, industry restraint against raw momentum, the economic promise of AI against the real, tangible cost of building it out. The live question is no longer whether oversight is coming. It is whether anyone can impose it fast enough for it to matter.
| The Brief |
01
Congress agreed on AI 417-3 — just not on the part Shapiro is asking for.
Pennsylvania Gov. Josh Shapiro used a Pittsburgh summit on Thursday to push for federal AI regulation, calling for independent oversight of frontier models along with transparency and safety requirements. He took a shot at President Trump for dismissing AI concerns as a "hoax" and urged House Speaker Mike Johnson to get something done before recess. Shapiro also said he wants Trump to raise "practical guardrails" when he meets China's President Xi next week. The House, for its part, had just cleared the one AI measure it could agree on: the Ratepayer Protection Act, passed 417-3, which makes large data centers pay for their own power infrastructure. Nothing comparable has moved on frontier models.
Why it matters: A Democratic governor publicly leaning on a Republican Speaker to move a bill before recess is not how this usually works, and it says the pressure on AI has stopped running along party lines. The contrast with that 417-3 vote is the tell: Congress can move almost unanimously on the part of AI that shows up on a constituent's power bill, and not at all on the part that decides how frontier models get tested before release. For anyone running AI inside a campaign or an agency, that means the standards you will actually be judged against keep arriving from governors and statehouses rather than Washington, one jurisdiction at a time. There is no single federal benchmark to build against, and nothing this week suggests one is close.
02
Anthropic says the honor code is finished. It also talks to the White House every day.
Anthropic policy chief Sarah Heck told a Washington summit that AI companies should not be left to self-regulate on an "honor code" and that the industry needs government oversight. Heck noted that Anthropic is in daily contact with the White House and has spent the year working with Congress. Her remarks followed a call from Anthropic CEO Dario Amodei for the industry to deliberately slow down model development, a position that Sam Altman at OpenAI, Elon Musk, and Demis Hassabis at DeepMind all quickly endorsed.
Why it matters: When a frontier lab argues in public against being trusted, mandatory oversight is usually close enough to plan around. Notice also who benefits: the position favors whichever labs are best placed to pass an audit, and Anthropic is telling you in the same breath that it has the regulator on the phone daily. Companies that welcome outside review get a hand in defining what review means. Rivals who resist it get cast as the reason the rules were necessary. For campaigns and public agencies buying these tools, the near-term consequence is that vendor vetting gets heavier, and that audit trails, model documentation, and disclosure move from selling points to conditions of purchase, set by regulators rather than by the vendors.
Presented by Vector Political
Most of the voters your canvassers went to see were never home.
Every field director knows the biggest column on the walk report: not home. Those are the voters the program was built to reach, and most of them stay unreached all cycle. Vector Political picks up where the knock left off. We open real, two-way deep-canvass conversations with the exact voters who did not answer the door, at scale, grounded in tested persuasion and backed by human review. The door stayed shut. The conversation still happens.
See how it works →
| Worth Knowing |
The 417-3 bill itself: Reps. Gabe Evans and Kathy Castor's Ratepayer Protection Act makes any AI data center drawing 100 megawatts or more pay for its own power infrastructure instead of handing the cost to consumers, which sounds decisive until you notice states can adopt the standard or ignore it and the Senate has not touched it. Read the story → (CNBC) |
Britain fast-tracked data centres by granting them Critical National Infrastructure status and got a revolt for it: campaigners have already overturned approvals in Buckinghamshire, US polling puts opposition to having one built nearby at 71%, and Democratic governors in New Jersey and Virginia ran on the issue in 2025 and won. See the backlash → (BBC) |
The real fight is preemption: whether Washington can stop Texas, Florida, and Utah from enforcing AI rules they have already written, with Speaker Mike Johnson resisting anything that slows development because of China and Hakeem Jeffries pressing Congress to act now, which leaves campaigns working around conflicting state rules on AI ads and chatbots until someone wins. Read the breakdown → (NPR) |