Written By | Torch Media Africa
The agencies — NSA, CISA, and the Treasury Department — will assess whether these "frontier models" pose cybersecurity or national security threats
Search
WASHINGTON, D.C.
It happened quietly on a Monday afternoon, June 2, 2026 — no fanfare, no televised signing ceremony. President Donald Trump signed an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security," and with that single act, he fundamentally altered the relationship between the U.S. government and the companies building the most transformative technology in human history.
The order asks AI developers to voluntarily hand over their most powerful, cutting-edge models to federal agencies for up to 30 days before public release. The agencies — NSA, CISA, and the Treasury Department — will assess whether these "frontier models" pose cybersecurity or national security threats. It is, in essence, a pre-publication review program for artificial intelligence. And it is unlike anything ever attempted by a U.S. president.
A Seismic Shift In AI Governance
The Biden administration's 2023 AI executive order required developers of large models to notify the government before training and share safety test results. But it never asked companies to share the models themselves with intelligence agencies.
Trump's order goes further. It is framed as pro-innovation and deregulatory, describing its goal as making the U.S. the global leader in AI development. But embedded within that rhetoric is a security framework that gives the national security apparatus an unprecedented window into the most powerful AI systems ever created.
The order asks companies to participate in a "classified benchmarking process." If a model crosses a certain capability threshold — determined by the NSA Director — it becomes a "covered frontier model" and falls under the voluntary 30-day review.
The Scrapped Order — And The Revised One
Just two weeks before signing, Trump scrapped a nearly identical version following intense pushback from AI industry lobbyists who argued that mandatory review would stifle innovation, raise trade secret concerns, and give federal agencies too much leverage.
The revised order made the framework explicitly voluntary. The legal text states it shall not be interpreted as authorizing "compulsory licensing, preclearance, permitting or government approval for the development or release of AI models."
But legal experts note that while participation is technically optional, companies that choose not to engage may find themselves at a disadvantage — particularly when bidding on federal contracts, seeking export licenses, or trying to stay in regulators' good graces. The word "voluntary" may carry less weight than it appears.
NIST, NSA, And The New Architecture Of AI Oversight
Within 60 days — by early August 2026 — the Treasury and Homeland Security Secretaries, acting through NSA and CISA, must design the voluntary framework. NIST plays a central coordinating role, consistent with its history of setting voluntary technical standards.
The secrecy element has raised concerns. If the benchmarks for determining whether a model is a "covered frontier model" are classified, then the public — including the companies developing AI — cannot know exactly what standards they are being evaluated against. Critics argue this creates opportunities for regulatory overreach or politically motivated targeting.
Silicon Valley Reacts — Cautiously
The reaction from major AI companies has been measured.
Company Response
Anthropic "Looked forward to working with the administration on implementation"
Google Described the initiative as "an important step"
OpenAI Called for frameworks developed "through democratic institutions, informed by technical expertise and broad stakeholder input"
The measured tone makes sense given the turbulent context. The week the order was signed coincided with Florida becoming the first state to sue OpenAI over ChatGPT safety, and Anthropic filing confidentially for an IPO reportedly valuing the company at approximately $900 billion.
Then there is the Mythos Preview incident. In April 2026, Anthropic voluntarily limited release of a new model after internal testing revealed an unusual ability to identify and exploit software vulnerabilities. That self-imposed restriction is widely believed to have helped accelerate the executive order.
A $920 Million AI Cloud Deal — And Questions Of Conflict
Adding another layer: Google signed a cloud services agreement with SpaceX for AI compute capacity worth approximately **$920 million per month** as of June 8, 2026. Critics argue that when a single cloud contract runs nearly $1 billion per month, the idea that a company would freely share its most proprietary technology with government reviewers out of civic spirit seems optimistic at best.
Expert Perspective: A Landmark, But Not A Straitjacket
Policy experts describe the order as a landmark document that stops well short of what many safety advocates would have liked.
"This is the most significant federal action on AI security since the Biden order, but it lacks the mandatory teeth that would give it real enforcement power," said one Washington-based AI policy researcher.
Others have taken a more optimistic view: "If the benchmarking process is well-designed and companies see genuine value in the government's security assessments, the voluntary nature could actually make it more effective than a mandate — because companies will participate proactively rather than resentfully."
What This Means For You
For ordinary Americans, the immediate impact is abstract. But the longer-term implications are not. The AI systems being evaluated under this framework are the same ones increasingly making decisions about job applications, medical diagnoses, financial loans, and criminal justice.
If the voluntary review process works, Americans could benefit from assurance that the most powerful AI models have been assessed by government security experts before deployment. If it fails, the real regulatory action will shift to the courts, state legislatures, and the international arena.
Europe has already moved much further down the mandatory regulation path with the EU AI Act, which came into force in 2024. The contrast between Washington's voluntary approach and Brussels' mandatory regime is increasingly stark.
The Bottom Line
The signing of this executive order marks a genuine inflection point. For the first time, the U.S. government has created a formal pathway through which NSA and CISA can engage directly with companies building the most powerful AI systems on Earth — systems capable of writing code, running autonomous tasks, and identifying software vulnerabilities.
Whether that pathway is used wisely, fairly, and effectively will define the next decade of American AI policy. The 30-day review window is now open. The clock is ticking.
