White House Chief of Staff Susie Wiles posted Wednesday that the Trump administration will not choose winners and losers in the AI race, a statement that arrived the same week the government announced pre-deployment model testing agreements with Google DeepMind, Microsoft, and xAI, and as the White House considers a working group to vet frontier models before public release.
The contradiction is worth sitting with. Wiles made the statement on X, the platform owned by Elon Musk, whose company xAI is one of the three firms that just signed pre-deployment evaluation agreements with the Commerce Department's Center for AI Standards and Innovation. OpenAI and Anthropic had earlier versions of those agreements in place from 2024, now renegotiated under the current administration. The government reviewing models before release is not neutral rulemaking. It gives established labs with compliance infrastructure and government relationships a structural advantage over startups that lack the resources to navigate that process. Saying you won't pick winners while formalising access arrangements with a select group of frontier labs is a policy position, not an absence of one.
The context around Anthropic makes the neutrality claim harder to sustain. The company's new model, Claude Mythos Preview, was capable enough at finding security vulnerabilities that Anthropic voluntarily restricted its release. That triggered White House attention, reportedly prompted discussions about a formal review executive order, and coincided with reports that the administration was considering deescalating its earlier feud with Anthropic. The EU separately reached out to Anthropic to test European banks for the vulnerabilities Mythos could identify. A policy architecture that responds to the capabilities of specific named companies, and adjusts its relationship with those companies accordingly, is making choices about who matters, regardless of the stated intention.
The export control picture reinforces the point. The administration has been active on restricting Chinese access to American AI models and compute, with an April memo from the White House Office of Science and Technology Policy directing industry to share information about unauthorised model extraction by Chinese developers. Export controls on chips, model weight restrictions, and procurement rules all advantage companies with existing government relationships and US-based infrastructure. Nvidia, Google, Microsoft, and the major frontier labs are positioned to benefit from those rules in ways that smaller labs and startups are not. That is not neutral market competition. It is industrial policy with a different label.
For SF readers, the relevant question is what "not picking winners" actually means in practice for fundraising, product development, and regulated launches. The honest reading is that the administration prefers market competition over direct subsidy or mandated winners, which is different from genuinely neutral policy. That distinction matters. Pre-deployment review agreements create compliance overhead that advantages well-resourced incumbents. Government procurement for AI tools concentrates revenue among a small number of established vendors. Safety standards, if formalised through an executive order working group stacked with large company executives, as Bloomberg's March report on Trump's technology council suggests is the default composition, will reflect incumbent preferences. Startups building in regulated categories such as healthcare AI, financial AI, or security applications need to account for those structural realities, not just the rhetorical framing.
The positive signal from Wiles's statement is that the administration is not moving toward direct nationalisation or mandated consolidation of AI labs, a scenario The Atlantic explored in April as a plausible outcome if US-China competition intensifies. Founders can treat the current environment as one where government will set boundaries around security and export risk, but will not dictate market structure within those boundaries. That is a workable environment for building, provided you understand which boundaries are hardening. Pre-deployment review for frontier models, tighter export controls on dual-use AI capabilities, and procurement rules favouring domestic providers are the areas where policy is becoming more prescriptive. Consumer applications, open-source development, and enterprise software outside regulated sectors remain largely unencumbered.
The practical read for founders is to watch the executive order working group composition and mandate closely. If it includes representation from small and mid-size AI companies, the resulting framework may genuinely reflect competitive neutrality. If it mirrors the technology advisory council Bloomberg described in March, dominated by established names and notably missing the leaders of the top AI labs, it will produce rules that serve incumbent interests while the administration maintains its "no winners" framing. Either way, the direction of travel is toward more government involvement in AI development, not less. Planning for compliance costs, pre-deployment disclosure, and security review as permanent features of the landscape is more useful than taking the neutrality statement at face value.","excerpt":"White House Chief of Staff Susie Wiles stated the Trump administration won't pick winners in AI, but simultaneously the government formalised pre-deployment testing agreements with Google DeepMind, Microsoft, and xAI, and is weighing a model review executive order triggered by Anthropic's Claude Mythos.
Also read: Anthropic's consumer Claude push shows even safety-focused labs cannot ignore ChatGPT's daily-use dominance • Users are already jailbreaking AI scam bots, turning fraud into a consumer defence arms race • White House pre-release AI model reviews would turn speed into a privilege incumbents can afford
