New OpenAI Hire Warns Government Monopoly on AI Could Lead to ‘Very Scary Outcomes’

Dean Ball, formerly a White House Senior Policy Advisor and now working on frontier AI policy at OpenAI, delivered a pointed warning about where U.S. AI oversight may be heading on Nathan Labenz’s The Cognitive Revolution podcast. Ball argues that the Trump administration’s move to classify AI testing procedures and centralize oversight within the intelligence community is setting up “a potentially very bad future where access to frontier models is gated.”

Ball wasn’t being political. Rather, he warns that government monopolization of frontier AI could lead the country to “very scary outcomes from a civil liberties perspective.”

A Non-Partisan Framing

Ball was careful to separate his critique from any party label. “If I didn’t know who the president was, if you didn’t know the party, didn’t know the name… and you told me about this, I would say I’m concerned about that,” he told Labenz. He extends that posture to prior administrations as well, noting that he criticized the Cyber Executive Order when it was signed because its voluntary pre-deployment testing program would be “primarily classified” and run by the intelligence community.

“Speedrunning” a 2023-Style Panic

According to Ball, what is unfolding in Washington looks familiar. He says the administration is “speedrunning” a regulatory-panic mentality from spring 2023, driven by roughly 15 people improvising AI governance without deep AI context. The “15 people” figure is his characterization, not an official headcount, and he uses it to underscore how concentrated the decision-making is.

The mechanism that worries him most is the migration of frontier-model evaluation away from civilian standards bodies and toward the national security apparatus. The federal AI build-out is enormous: the Department of War’s FY 2027 budget request includes $58.5 billion for Artificial Intelligence and Combined Joint All-Domain Command and Control, with $46.0 billion earmarked for a multi-year mandatory investment in a sovereign AI Arsenal and the deployment of frontier AI models through the Department’s enterprise GenAI.mil platform. Ball’s argument is that when an oversight regime sits inside agencies of that scale and secrecy, the line between regulator and customer can blur.

The Anthropic Contract Claim

Ball reports that the Department of War appears to be winding down Anthropic contracts, potentially reaching zero by the end of the year, while other federal agencies continue using Anthropic without issue. He frames this as a data point worth watching.

He also noted that the NSA, technically part of the Department of War, reportedly honored Anthropic’s red lines around “domestic mass surveillance and autonomous lethal weapons,” which he reads as a sign that “a little bit of ambiguity is… an intrinsic part of this whole process.”

Why It Matters for the Broader AI Industry

Ball’s concern is that testing and access to the most advanced AI models could gradually become concentrated inside classified government channels. In his view, that raises civil liberties questions and increases the risk that access to frontier AI becomes controlled by a small number of government institutions.

The issue also has implications for the broader AI industry. Companies are spending hundreds of billions of dollars on the assumption that advanced AI tools will be widely deployed throughout the economy. If access becomes more restricted, the path to adoption and monetization could look very different.

Ball argues that investors and policymakers should pay close attention to who controls the testing process and who ultimately gains access to frontier models.

Contact [email protected] for any questions or corrections.