The US government has launched the clearinghouse "Gold Eagle" as part of the AI decree. Cyber vulnerabilities are to be addressed here in a coordinated manner.
The Trump administration is continuing its AI course in cybersecurity. As the White House announced, Gold Eagle, the first operational program from the AI decree signed in early June, has been launched. The clearinghouse is intended to provide a common platform for government agencies and private companies such as “open-source software partners” and operators of critical infrastructure to jointly address cybersecurity vulnerabilities.
The program is based on Executive Order 14409, which President Donald Trump signed on June 2, 2026, under the title “Promoting Advanced Artificial Intelligence Innovation and Security”. Gold Eagle is intended to act as a “force multiplier” – bundling existing capacities of the state and economy rather than having them work in parallel. According to the White House, the goal is to avoid duplicate scanning efforts and instead to specifically pass validated vulnerabilities to teams in industry and government.
In addition to the White House, the US Department of the Treasury, the Department of Homeland Security through its cybersecurity agency CISA, and the Pentagon are involved. Treasury Secretary Scott Bessent stated, according to the White House announcement, that his department “is working hand in hand with the private sector to safeguard our financial institutions, close vulnerabilities, and protect the integrity of the U.S. financial system”.
Videos by heise
Context: Export controls and European backlash
Gold Eagle does not come out of thin air. The US government has been trying to gain control over the AI models of major US providers for some time -- usually on the grounds of “cybersecurity concerns”. Just in mid-June, the US government ordered Anthropic to block access to its frontier models Mythos 5 and Fable 5 worldwide – including for foreigners within the US. The trigger was a discovered jailbreak that could bypass security mechanisms and make the powerful cybersecurity functions of Mythos 5 usable without restrictions. The US government feared that the consumer product Fable 5 could thus become an uncontrollable cyber weapon. At the end of June, the Department of Commerce lifted the blocks after more than two weeks.
The government proceeded similarly with OpenAI: its new AI model GPT-5.6 with the most powerful variant “Sol” was made accessible to only a small circle of trusted partners from the outset – at the request of US authorities. Unlike with Anthropic, where the model had to be released and then withdrawn, OpenAI took the path of controlled release from the beginning.
The interim blocks caused concern in Europe. The EU Commission has since been working on an action plan that will develop concrete emergency measures by the end of the year in case a third country restricts access to security-relevant AI. EU Digital Commissioner Henna Virkkunen emphasized that Europe cannot “rely solely on non-European solutions for capabilities that are crucial to our security”. The Commission intends to develop the guidelines together with the EU cybersecurity agency Enisa and build a testing platform for AI models.
(rie)