admin
President Donald Trump quietly signed an executive order that asks big AI companies to let the government take a look at certain advanced “frontier” models before they go public. The order sets up a voluntary prerelease review of up to 30 days, a sharp narrowing from an earlier 90‑day draft. The move came without a public ceremony — no cameras, no fanfare — just a scaled‑back deal meant to straddle national security and tech competitiveness.
The new, voluntary prerelease plan
The order tells federal agencies to build a voluntary prerelease process for covered frontier models and to create cybersecurity benchmarks. Companies would be asked to provide early access up to 30 days before a public launch. The text is careful to say this is not a mandatory licensing or preclearance regime. Treasury Secretary Scott Bessent and Vice President JD Vance are among officials involved as the White House rolls out the plan. The administration also calls for a “cybersecurity clearinghouse” to share fixes and warnings about AI‑linked vulnerabilities.
A narrow compromise shaped by industry lobbying
Reports say tech leaders pushed back hard, and they got what they wanted: a weaker, voluntary framework and a shorter review window. CEOs like Sam Altman and Dario Amodei were part of the conversations that clearly influenced the final language. Translation: Big Tech lobbied, the White House listened, and the end result looks more like a gentleman’s agreement than a serious safety regime. If you were expecting muscle, you got a handshake and a polite request.
Security claims, but real teeth are missing
The order speaks in national‑security tones and orders agencies to harden critical systems. That’s fine in theory. In practice, 30 days to inspect a complex, modern AI model — much of which is opaque — is a sliver of time. Benchmarks and a clearinghouse sound useful, but the directive leaves lots of implementation questions open. Which agencies will run the tests? What happens if a lab refuses? The White House hints procurement penalties could be used, but the EO stopped short of clear enforcement tools.
Bottom line: a start, not a solution
This executive order is a step toward taking AI risks seriously, and the national‑security framing is the right tone. But calling it a victory for oversight is generous. It’s a compromise that placates industry and avoids heavy regulation. Conservatives who worry about both security and economic strength should want more clarity — not private signings and vague promises. If the government truly means to protect Americans and preserve U.S. AI leadership, the next moves must include clear benchmarks, transparent results, and real consequences for labs that decline to cooperate.
