OpenAI's midterm plan: 'AI speed limits' and 'compute campuses'
Two days before the US midterms, OpenAI has shown its hand for the morning after: a policy and PR campaign built around mandatory safety rules for frontier models, and a new name for the buildings it puts them in.
OpenAI will launch the campaign immediately after Tuesday's vote, with four proposals: mandatory safeguards for the most powerful AI systems, which it is calling "speed limits"; community compacts and model state legislation for data centers; a worker-disruption agenda; and a push for US-shaped global standards — Axios reported Sunday. The most striking item is the first one, because it is a reversal. In September OpenAI was asking Congress whether an industry-wide slowdown would even be legal; now it says it will campaign for one, under a name that sounds like traffic law rather than industrial policy.
Chris Lehane, OpenAI's chief global affairs officer, framed the timing as a transition rather than a reaction. Tuesday is "an electoral portal from one era to the next," he told Axios, and the choice for a company in OpenAI's position is to "try to chase the moment, or seek to inform the moment via the power of ideas." On the substance: "You're going to see us talk more and more about the concept of speed limits. Our belief is people should always remain in control of this technology." Lehane's team has been stockpiling the proposals since August, and the rollout is timed to the lame-duck session that opens November 9, when the current Congress returns for the awkward weeks before new members are sworn in on January 3.
Why now, and why this
The arithmetic is public. Democrats are favoured to take the House, which turns a White House that has spent two years saying yes into one half of a divided government that will spend two years saying prove it. Every tech chief executive who spent the past two years close to Trump becomes a subpoena target the moment the gavels change hands. And the 2028 presidential field starts forming on Wednesday, which is why Lehane describes the midterms as "a bit of a movie trailer" and says flatly that "AI is going to be the issue in 2028." Meta is making the same bet in its own way, assembling a bipartisan influence operation, as Axios reported Saturday.
The speed-limits idea is not new for OpenAI — it builds on the standards proposal the company published earlier this year — but the politics of it are. In September, OpenAI asked members of Congress for guidance on whether orchestrating an industry-wide slowdown would run into antitrust law, which we covered at the time — OpenAI asked Congress if an industry-wide slowdown is even legal. Two months later the company is volunteering to write the slowdown. Read the two positions together and the direction of travel is clear: OpenAI has decided it would rather help design the constraint than be handed one, and it has found a metaphor that makes constraint sound like a safety feature instead of a brake.
What is missing is everything that would make "speed limits" a policy. A speed limit has a number, a road, and someone with a ticket book. OpenAI has not said what the threshold is, whether it applies to training runs or releases, who sets it, or which regulator enforces it. Until one of those exists, this is a phrase that poll-tests well and commits the company to nothing — which is also exactly why it can be announced before an election nobody has won yet.
The rebrand is the cheap part
The data center plank is more specific and, in the short run, more consequential. OpenAI is promising community compacts and commitments to smooth approvals, plus a proposal for state legislation that would give residents a meaningful say in how projects are built and what they get out of them. It is also arguing that what it builds should be called an "AI compute campus" rather than a data center, on the grounds that the campuses transform host communities economically — pointing to its Ports Pike project in Pike County, in Appalachian Ohio, and the thousands of jobs it says the project will bring.
Lehane's read on the backlash is that it is healthy: "Communities should hold companies responsible and accountable so they really and truly get a transformative benefit from these projects."
A rebrand is the cheapest item on the list, and it is the one people will remember, which is a problem. The opposition is not a vocabulary failure. Sixteen states have at least one gubernatorial candidate backing a data center moratorium heading into Tuesday's vote — Data center moratoriums are on the ballot in 16 governor races — and the complaints are about water, power, rates and process, none of which changes when the sign outside changes. Local negotiators have been asking for the opposite of a rebrand: Amazon said in October it would stop using non-disclosure agreements in negotiations with local governments, TechCrunch reported, after years of projects arriving as fait accomplis. A campus that still signs an NDA is a data center with a brochure.
The framework OpenAI's speed limits would have to plug into is itself unpublished, and the administration says it does not have to hand over the draft. In a letter, the White House argues it need not share the full draft of its voluntary AI testing framework, citing national security and commercial exemptions, Leo Schwartz reported at The Information on Sunday.
The background makes the refusal sharper. A June executive order asks labs to submit their most advanced models for government testing up to 30 days before public release — a window cut from 90 days after industry lobbying — and set an August 1 deadline for the framework that governs the tests. The framework was finished in early August. It has not been published. Neither the administration nor the participating companies is obliged to release results or even confirm who took part, and the Center for Democracy and Technology wrote to the White House in September asking it to publish the document. Sam Altman has said OpenAI submitted its Astra model for review and described the process as productive, which is the only public account anyone has of how it works.
That is the awkward join between the two stories. OpenAI wants to legislate speed limits for frontier models in a building where the existing voluntary test regime has secret criteria, secret participants and secret results. The exemptions cited in the letter are the interesting part: a commercial exemption covers whatever the labs submit, a national security one covers whatever the government wrote, and if both hold there is nothing left for a reader, a state regulator or a rival lab to check. A voluntary regime can survive being unenforceable. It cannot survive being unverifiable, because then it is a press release with a government logo on it.
What to watch: whether the CDT or another group litigates for the framework, whether the first AI hearings of the new Congress subpoena it, and whether OpenAI's speed-limit proposal ever contains a number.
If a lab writes the safety rule it will be judged by, does that count as regulation — or is it a press strategy with a deadline? Write to us via our contact page.
