Trump grants exclusive access to new AI before anyone else

Contents

After some back and forth, US President Donald Trump finally published an executive order on artificial intelligence on Tuesday. It establishes a whole host of new working groups dedicated to topics surrounding AI and IT security. A gaping hole remains in the security of AI itself – Trump announced a corresponding agreement between the US government and the AI industry on the first day of his second term.

While he declares the promotion of AI innovation and security to be an official goal, he then orders no concrete measures to strengthen AI security. This is to be achieved through “collaboratively with the private sector to modernize government and private sector information systems and harden them against external threats.” In reality, insiders fundamentally question the greater threat, and more recently, risks from internally developed AI agents have been added.

Trump limits himself to general orders to “prioritize” IT defense for the military, intelligence agencies, civilian authorities, and their service providers – within 30 days. Federal programs not further specified for “enhance AI-enabled defensive tools” are to be expanded or established. Incidentally, the Attorney General is to take stronger action against criminals who use AI for their schemes.

Shopping spree without a budget

And Trump urges authorities to spend money: On the one hand, IT security tools and services are to be procured for federal, state, and local government agencies, as well as for operators of critical infrastructure “such as rural hospitals, community banks, and local utilities.” Who is supposed to take care of this and from which budget it should be covered, Trump leaves open. In this respect, it is more of a wish letter to parliament, which determines the federal budget.

On the other hand, the White House budget office is to search for existing funding programs whose funds can be dedicated to subsidizing AI development: specifically advanced AI for discovering security vulnerabilities. One or another reallocation may also be permissible without a parliamentary resolution, but in total it will not achieve much.

Intelligence agencies and the military are to find security vulnerabilities

In parallel, the Department of the Treasury, together with the Secretary of Defense represented by the NSA intelligence agency and the Secretary of Homeland Security represented by CISA (Cybersecurity and Infrastructure Security Agency), is to establish an AI clearinghouse. This is intended to coordinate and de-escalate the scanning for software security vulnerabilities in “voluntary collaboration with the AI industry and operators of critical infrastructure.”

The clearinghouse is also to find and confirm such vulnerabilities itself, as well as coordinate and prioritize their closure, including the distribution of security updates. The executive order does not mandate public disclosure. In the end, this means that the NSA and others will search for, find, and then control security vulnerabilities and decide if and when they will be fixed and by whom. There is no mention of support for the struggling vulnerability database NVD of NIST.

Since AI cannot do everything alone, and Elon Musk's Doge has fired numerous IT experts without good reason, new specialists are needed. To this end, the US government proclaimed the “US Tech Force” at the end of 2025. This long-term program allows nearly 30 AI corporations to place approximately 1,000 of their employees in US ministries for two years each, who pay exceptionally high salaries in return. Most of the participating corporations have donated directly or indirectly to Donald Trump. His current executive order provides for the expansion of the US Tech Force: Authorities require more IT security experts.

Exclusive access to the latest AI

Section 3 of the executive order aims for exclusive priority for the US government regarding so-called “frontier models.” This refers to the most advanced large AI models, i.e., the latest state-of-the-art. They are to be made exclusively available to the government for 30 days first, rather than coming directly to market.

Even after that, free market access is not desired. Instead, operators and the government are to jointly determine which trusted partners may use the frontier model “to promote secure innovation and strengthen the cybersecurity of critical infrastructure.” The President cannot enforce these restrictions, which is why he is establishing a voluntary framework for AI developers. They are thus “voluntold” - appointed as volunteers.

The working group known from the clearinghouse, consisting of the NSA, CISA, and the Treasury Department, is again responsible. To determine what qualifies as a frontier model, they are to develop a secret benchmark. The NSA will use this to evaluate new AI models. If it classifies one as a frontier model, the NSA will inform AI developers and researchers, “as appropriate.”

Participants in the framework may also inquire whether an AI model they are currently working on is likely to be classified as a frontier model. Otherwise, a corporation might underestimate its technical achievement, violate its voluntary commitment, and simply release the super-AI.

• Regarding the executive order “Promoting Advanced Artificial Intelligence Innovation and Security”

(ds)