The drafter of America's AI strategy is now writing policy for OpenAI

When the person who would have been the most natural adversarial counterpart for federal AI policy becomes an employee of the firm that policy targets, the regulatory game does not reset by legislation. It resets by a job announcement. That is the structural read of Dean Ball's move from the White House Office of Science and Technology Policy to OpenAI, disclosed on this week's Cognitive Revolution podcast with host Nathan Labenz.

Ball was not a bystander to US AI policy. As a senior fellow at the Foundation for American Innovation, a think tank that has shaped much of the Trump-era technology agenda, and a staff drafter inside the White House Office of Science and Technology Policy, he helped write the AI Action Plan, the White House's 2025 strategy for keeping American AI ahead of competitors while managing safety, energy use, and adoption inside the federal government. He also authors the Hyperdimensional Substack, where he has been one of the more pointed outside voices on governance of the most capable AI models. On the podcast, Ball frames the AI Action Plan as roughly "30 to 40 percent" finished, with substantial work on export controls, federal AI procurement, and state-level preemption still ahead.

Ball says he is taking that unfinished work to OpenAI, to be done on the company's behalf. The move is part of a wider pattern in which policy talent migrates from the White House and adjacent think tanks into frontier AI labs, the small group of companies building the most capable models. The result is that the people best positioned to design oversight of frontier AI are now employed by the firms that oversight would constrain.

The structural consequence is concrete. Federal AI policy in 2026 and 2027 will turn on a handful of contested rulemakings: export controls on advanced AI chips and models, federal procurement rules for AI systems used by the military and civilian agencies, state-level AI laws and the question of whether federal policy preempts them, and energy and grid siting for the data centers training and running frontier models, much of which now flows through the Federal Energy Regulatory Commission. In each of these, the institutional knowledge, deliberative tempo, and rulemaking instincts that would have produced adversarial comment letters, competing analyses, and slow internal review have moved from the Office of Science and Technology Policy and the Foundation for American Innovation into OpenAI.

Ball is candid about what he is taking with him. On the podcast he lists export controls, government AI use, state AI regulation, and emerging capability thresholds in coding, cyber, and robotics as the areas where he expects to focus. These are the same topics the AI Action Plan named as priorities. The list is also a near-complete map of the rulemakings that would have most affected OpenAI's deployment decisions, internal capability releases, and the recursive self-improvement thresholds, the points where a model improves itself, where the lab's own internal review currently takes the place of external review.

That is the part that should make readers who track these dockets pay attention. The structural move is not a personnel story. It is a unilateral commitment in the policy game: the actor who helped design the oversight framework is now inside the entity it regulates, and the regulatory counterparty has been relocated rather than reformed. No statute changed. No agency moved. The negotiating table shifted.

Ball's own retrospective on the AI Action Plan helps calibrate the stakes. He credits the plan with concrete wins: pairing AI growth with nuclear energy commitments, pushing the Federal Energy Regulatory Commission toward faster grid interconnection for data centers, and accelerating military AI adoption. He treats export controls and state preemption as unfinished. The critique, which he largely accepts on the podcast, is that the plan's success depended on the same small group of staffers and outside experts who are now being hired by the companies they were supposed to oversee.

The legitimate concerns are easy to name. There is the revolving door between White House AI policy and the lab being regulated, the in-housing of policy capacity by a small number of frontier labs, and the risk that future AI policy gets drafted inside the companies that have to live with it. None of these is a new problem in Washington, but the speed and the concentration are new. Three or four frontier labs now employ a significant share of the people who would have been the most credentialed outside critics of those labs' deployment choices a year ago.

The constructive frame is not that the move is bad or good in itself. It is that the shift is visible and actable. State legislatures, federal comment periods on export-control and procurement rules, and Federal Energy Regulatory Commission dockets on data-center interconnection all remain open. Readers who care about how AI is governed can show up at those points. The question is whether the public-interest counterweight will be staffed at the same depth as the in-house policy teams now being built on the other side of the table.

What to watch next is specific. The Office of Science and Technology Policy will need to backfill Ball's role or shift the work elsewhere. The State Department's export-control rulemaking on advanced AI models is the next live federal action. State legislatures, particularly in California, Texas, and New York, will test the federal preemption question. And OpenAI's deployment decisions on coding agents, cyber capabilities, and robotics, the threshold areas Ball named on the podcast, are the places where the in-housing of policy capacity will show up fastest in the real world.