Trump quietly opened state access to the most powerful AI models: where does security end and control begin

The White House changes its tone: artificial intelligence is no longer just a market

Donald Trump signed an executive order on artificial intelligence on June 2, 2026 — without much ceremony, but with consequences that may be felt not only by American tech companies. The White House document received the official title “Promoting Advanced Artificial Intelligence Innovation and Security” and formally speaks about supporting innovation, protecting American leadership, and strengthening cybersecurity.

But behind the calm language of the order lies a harsher reality: the state wants to get an early look at the most powerful AI models even before they hit the market or reach a wide range of partners.

This is not called licensing.

This is exactly what the White House emphasizes: companies are not required to undergo classic approval for model release. However, the document introduces a voluntary scheme whereby developers of frontier models can submit models to the state for preliminary evaluation for up to 30 days. The Verge describes this as a new voluntary mechanism that gives federal structures access to advanced models before public release to assess risks to cybersecurity and critical infrastructure.

For the market, this looks softer than a direct ban.

For the companies themselves — not necessarily.

When it comes to the strongest models, closed evaluation criteria, intelligence agencies, financial regulators, and infrastructure, the word “voluntary” begins to sound different. Especially if refusal to cooperate can spoil relations with the state, the defense sector, banks, or large corporate clients.

Why the order appeared now

Not long ago, the Trump administration tried to demonstrate the most free approach to artificial intelligence. The main line was clear: America should not lose to China due to bureaucracy, restrictions, and fear of its own tech companies.

Now the tone has changed.

The Verge writes that the new version of the order became a compromise after disputes within the administration. A tougher version, according to the publication, was previously postponed due to fears that it could hit American leadership in AI. In the final document, the period of preliminary access was reduced to 30 days, and the mechanism itself was framed as voluntary cooperation, not as mandatory release permission.

The change in rhetoric did not occur due to a philosophical debate about AI ethics.

The main irritant is cybersecurity. American authorities have seen that advanced models are already capable of not just writing texts or code, but finding vulnerabilities in complex software, helping analyze old systems, and potentially accelerating the work of those looking for weak spots in banks, operating systems, browsers, and infrastructure.

Claude Mythos became an alarming signal for banks and regulators

One of the reasons for the sharp attention was the launch of Anthropic Mythos in the spring of 2026. Reuters reported on April 20 that the appearance of this model triggered a race in the banking sector: financial companies tried to gain access to the technology, and regulators began to assess what risks it poses to cybersecurity and the readiness of financial organizations.

Separate Reuters reports indicated that Asian financial regulators are also monitoring the risks around Mythos, as the model has powerful coding capabilities and potentially unprecedented ability to find cyber vulnerabilities.

This is an important detail.

In public policy, AI is often discussed as a threat to jobs, copyright, education, or media. But for the White House, the problem suddenly became much more concrete: what to do if a commercial model gains the ability to see weak spots in global software faster than many security teams?

Such a tool can be a shield.

But it can also become a weapon.

Where is the line between cooperation and oversight

The White House order talks about cooperation with the private sector, protecting intellectual property, maintaining US leadership, and strengthening systems against external threats. This is language that should reassure the industry: the state is not coming to break the market, it supposedly helps prepare for new risks.

But the very construction creates a new dependency.

If a model is recognized as powerful enough to fall into the category of covered frontier model, its developer is effectively offered early interaction with the state. Structures related to national security, cyber defense, and financial stability may participate in the process. The Times also writes that the final scheme provides for a classified benchmarking process and up to 30 days of evaluation before public release.

Formally, this is not a ban.

But for companies working with defense contracts, banks, cloud infrastructure, and government customers, such a system can become a new political filter. Not because the document explicitly states “allow” or “not allow.” But because in the real world, access to the market often depends not only on the law but also on the trust of the state.

Why it is important for Israel to follow this dispute in the US

For Israel, the American AI order is not a distant internal story of Washington. The Israeli economy, defense industry, cyber sector, and startup ecosystem are deeply connected with the American market, investments, clouds, chips, military technologies, and security standards.

If the US changes the rules regarding frontier models, it sooner or later affects allies.

Israel lives in a reality where cyberattacks, drones, missiles, intelligence, finance, civil infrastructure, and military solutions have long become part of a single threat system. After October 7, Israelis especially understand well: technological superiority does not work by itself. It needs to be tested, protected, and constantly adapted to an opponent who learns quickly.

For НАновости — News of Israel | Nikk.Agency this topic is important precisely as part of a new security architecture. Artificial intelligence can no longer be considered only as a business tool or convenient technology. It is becoming a factor of war, intelligence, bank protection, city defense, infrastructure resilience, and political control.

Anthropic and the conflict with the Pentagon intensify the political context

The story of Anthropic makes the situation even more sensitive. Reuters reported in March 2026 that the Pentagon recognized Anthropic as a risk to the supply chain, which limited the use of the company’s technologies by contractors of the American military department.

Anthropic itself then publicly stated that it received the corresponding letter from the Department of War and disagrees with such an assessment. The company wrote that it intends to seek a review of the decision, as it considers it wrong and harmful to US national security.

Against this background, any new “voluntary” interaction mechanism between AI companies and the state looks not just like a technical procedure.

It becomes part of a broader conflict: who controls the strongest models, who decides where they are safe, who gets early access, and can the state use cybersecurity as a tool to pressure inconvenient developers.

There is no evidence that the new order is already being applied against a specific company.

But the framework is created.

The main conclusion: AI enters the national security zone

The most important meaning of Trump’s order is not that the White House suddenly introduced strict AI regulation. He did something else: he moved the most powerful models from ordinary market logic into the realm of national security.

This is a more subtle and therefore more important turn.

While AI wrote texts, created images, helped programmers, and accelerated office work, it could be discussed as a matter of innovation and competition. When models begin to find vulnerabilities, analyze critical systems, help in cyber operations, and potentially influence the security of banks, energy, transport, and defense, the state is no longer ready to stand aside.

For Israel, Ukraine, and other countries living next to real war, this signal is especially clear.

AI is becoming not only the technology of the future. It is already turning into a tool of power, intelligence, protection, and pressure. Therefore, the main question now sounds not like this: will the state intervene in AI. It is already intervening.

The question is different: who will set the rules, how transparent they will be, and will protection against threats turn into a convenient mechanism for controlling those who create the strongest technologies of the 21st century.