Trump team adopts parts of Biden AI plan
- Trump’s AI team is not really reviving Biden’s whole playbook. It is keeping the machinery that tests powerful models, while scrapping the broader “safety-first” branding. - The clearest sign came on May 5, when Google DeepMind, Microsoft, and xAI agreed to let Commerce’s CAISI test frontier models before release. - That matters because Trump revoked Biden’s 2023 AI order in January 2025, but some federal oversight tools survived and quietly expanded anyway.
Artificial intelligence policy in Washington now looks less like a clean break and more like a relabeling job. Trump came in promising to tear out Biden’s AI guardrails. He did revoke Biden’s big 2023 executive order on January 20, 2025. But the part of the federal government that actually tests advanced models, writes standards, and coordinates agencies did not disappear. It got renamed, narrowed, and then used again. ### What is the actual news here? The immediate news is that the Commerce Department’s Center for AI Standards and Innovation — CAISI — is now evaluating frontier models from Google DeepMind, Microsoft, and xAI before those systems go public. That is a real federal gatekeeping role, even if the administration avoids talking about “AI safety” the way Biden did. It is also much closer to the old Biden-era model-testing idea than Trump’s day-one rhetoric suggested. (whitehouse.gov) ### What changed from Biden to Trump? The branding and the legal wrapper changed first. Biden built policy around risk management, civil-rights harms, and agency reporting. Trump revoked that flagship order, then replaced parts of the federal-use guidance with a more adoption-first approach in April 2025. The new OMB memo says agencies should focus on innovation, governance, and public trust — and it rescinds Biden’s M-24-10 memo rather than simply extending it. (politico.com) ### So what stayed? The technical backbone stayed. NIST’s AI Safety Institute was re-established in June 2025 as CAISI, but its core jobs still include voluntary agreements with model developers, evaluations of national-security risks, and coordination with agencies like Defense, Energy, Homeland Security, OSTP, and the intelligence community. That is basically the plumbing of AI oversight — less slogan, more machinery. ### Why keep that part? (whitehouse.gov) Because frontier AI got harder to ignore. CAISI’s own mandate centers on risks like cyber, bio, and chemical misuse. The White House’s March 20, 2026 legislative framework also said national-security agencies need enough technical capacity to understand frontier model capabilities and mitigate concerns in consultation with model developers. In plain English — even a pro-deregulation White House decided somebody in government still has to kick the tires on the most powerful systems. (nist.gov) ### Is this the same as Biden’s approach? Not really. Biden’s model was broader and more process-heavy across the whole executive branch. Trump’s is narrower, more national-security focused, and more openly pro-industry. The April 2025 procurement and governance guidance cut reporting burdens, redefined Chief AI Officers as adoption advocates, and created one “high-impact AI” category instead of a larger compliance structure. So the administration kept some guardrails, but only the ones it sees as compatible with speed. (nist.gov) ### Why do the company agreements matter so much? Because they show this is not just a think-piece policy. When Google DeepMind, Microsoft, and xAI signed on, the federal government gained early access to unreleased frontier systems for testing. That turns CAISI from a standards shop into an operational checkpoint. It is voluntary, not a licensing regime. But it still gives Washington a closer look at dangerous capabilities before launch. (whitehouse.gov) ### What does this say about Trump’s wider AI strategy? It says the strategy is “green lights for deployment, but not total blindness.” Trump’s July 2025 AI Action Plan pushed infrastructure, exports, deregulation, and “winning the AI race.” But that same plan also backed interpretability, robustness, evaluations, and federal adoption rules. Turns out the administration wants both acceleration and a small set of hard-edged checks around the frontier. (politico.com) ### What’s the bottom line? The Trump team did not adopt Biden’s AI philosophy. But it did keep some of Biden’s most practical tools — model testing, standards work, and interagency coordination — because governing frontier AI without them started to look reckless. (whitehouse.gov 1) (whitehouse.gov 2)