"Now, it’s Congress’s turn. We must address catastrophic risk without ceding ground to China or restricting Americans’ free expression," Senator Ted Cruz posted on X.
President Donald Trump’s 2 June executive order
On 2 June, President Donald Trump issued an executive order asking artificial intelligence companies to share their most powerful new models voluntarily with the US government for up to 30 days before wider release. The decision followed a "bruising tussle" inside the administration, where officials had debated the approach for months. The order is a presidential directive to government departments and agencies — not a law or regulation — and therefore cannot, by itself, compel private companies to act in a way that would necessarily stand up in court.
The 30‑day window and David Sacks’ influence
The order gives agencies an exclusive access period that lasts up to 30 days. That window was cut from an earlier 90‑day proposal after lobbying from Silicon Valley, represented primarily by venture capitalist David Sacks, who until March served as the administration’s AI czar but "clearly still has enormous sway." Critics in the piece question whether 30 days is adequate for complex threats: "If a model has serious bioterror risks, is the US intelligence community going to have time to analyse and assess within 30 days?"
No 'review', 'assess' or 'vet': the order’s explicit limits
The order does not say the government will check new models to ensure they are safe or free of unacceptable security risks. The text, as described, omits verbs such as "review," "assess," "oversee" or "vet" for what the government will do during the exclusive access period. It also specifically states it is not about "mandatory governmental licensing, preclearance, or permitting requirement" of new models. In practical terms, the order "sets expectations" on industry behavior but does not create an enforceable vetting regime.
Anthropic’s Mythos, staged release, and the frontier risk
The order mirrors the approach taken by Anthropic with its Mythos model: holding a powerful system back from general release and sharing it quickly with the US government. That precedent mattered in practice — the piece notes it took seven weeks for some parties to gain access during the Mythos rollout. But the author warns that not all models will resemble Mythos. Future "frontier models" might be capable of designing bioweapons or discovering whole new categories of software vulnerabilities. The order articulates no constraints on subsequent general release, and the article warns that "what happens when a model is actually too powerful and therefore too dangerous to release, and there are no safeguards or mitigations that can remedy that risk in any reasonable timeframe?" remains unanswered.
What this means for frontier labs, the US intelligence community, and Australia
- Frontier labs: The piece says labs are themselves worried and cites Anthropic’s decision on Mythos as a concrete example. But those labs face a "race dynamic" — commercial competition and strategic pressure against China — and are under pressure from "trillions of dollars in investment" to move faster, creating conflicting incentives.
- The US intelligence community: Early, exclusive access is designed to give agencies time "to figure out their own responses — including, presumably, how they might use it themselves against adversaries." The article questions whether the shortened 30‑day window will be long enough to assess high‑consequence risks such as bioterrorism.
- Australia and allies: The author reports that the Australian government and select Australian companies are "finally getting access" to new models under the framework, and that staged sharing may "smooth the process for US allies such as Australia to gain access to new models more quickly than the seven weeks it took with Mythos." At the same time, the piece notes the United States will likely hold back its most formidable capabilities for itself "as it does with most military‑grade technology."
Trust, regulation and the unresolved question
The order encourages a collaborative, non‑ordinary‑product framing for frontier models — a useful shift away from treating them like consumer products anyone can access with a subscription. But it remains voluntary and nonbinding. Australian AI Minister Andrew Charlton is cited warning that trust is necessary if Australia is to build a "world‑competitive industry" in the technology; the article concludes that trust "needs to be earned," especially when everyday items like kids' toys and shop‑made sandwiches face heavier burdens than frontier models. The central, concrete gap left by the order is procedural and legal: there is no verb in the directive that commits the government to reviewing or vetting models during the 30 days, and there is no mechanism in the order to stop a developer from releasing a model that others might judge "too dangerous to release."
The June 2 executive order nudges cooperation between industry and government, but as the piece makes plain, it is not regulation. Whether that nudge will be enough — or whether Congress, intelligence agencies, allies such as Australia, or the companies themselves will close the gap when a truly high‑risk model appears — is the unresolved question the order leaves on the table.
