A group of five Democratic senators has criticized the Trump administration's handling of artificial intelligence security, asserting that its inconsistent and opaque approach could inadvertently bolster Chinese AI alternatives and introduce new security vulnerabilities. The senators, including Kristen Gillibrand of New York, Adam Schiff of California, Mark Warner of Virginia, Chris Coons of Delaware, and Mark Kelly of Arizona, conveyed their concerns in a letter to top administration officials on Monday, August 4, 2026.
The lawmakers highlighted two specific incidents as examples of the administration's fluctuating strategy: an alleged "Hugging Face hack" involving OpenAI models and the Commerce Department's suspension of access to Anthropic's Fable 5 and Mythos 5 models for foreign nationals. They argued that the administration has been either too passive or overly interventionist, creating an unpredictable environment that undermines U.S. competitiveness.
Regarding the Hugging Face incident, the senators stated that OpenAI models "escaped testing," suggesting a lack of federal oversight. They emphasized that the government "cannot be passive as these capabilities emerge." During the period when Hugging Face was reportedly breached, the company allegedly had to rely on a Chinese open-weight model due to restrictions on U.S. "frontier models."
In the case of Anthropic, the Commerce Department reportedly used an "infrequently used authority" in June to direct the company to suspend all access to its Fable 5 and Mythos 5 models for foreign nationals, including those employed within the United States. The stated reason was an undisclosed national security concern, later described as a "narrow jailbreak finding." Because Anthropic could not immediately verify users' nationalities, it was compelled to disable both models for all users. The senators noted that the administration and Anthropic engaged in 18 days of closed-door negotiations before reaching an agreement.
The senators contended that while the administration's interventions might stem from legitimate security concerns, the lack of transparency in its standards and decision-making processes creates broader harm. They specifically cited the importance of keeping Congress fully informed when the Executive Branch exercises authority delegated from Congress, such as in export control administration.
They further warned that if American AI models are perceived as subject to sudden access disruptions due to "black-box U.S. Government processes," or as unreliable because U.S. AI labs are "overcorrecting" in response, both domestic and international entities may opt for Chinese or other foreign models. This outcome, they argued, would erode U.S. technological leadership and increase exposure to systems potentially carrying risks of censorship, espionage, intellectual property theft, and other supply chain security risks directed by the People's Republic of China.
The senators pointed to a tangible consequence of these actions, noting that the stock price of an entity-listed Chinese lab nearly doubled during the period Anthropic was under export controls.
Their letter, addressed to leaders in the White House, the Office of the National Cyber Director, and the departments of State, Treasury, and Commerce, requested clarification on several points. These included the standards the administration employs to assess national security risks posed by "frontier models," the legal authorities it intends to use for imposing restrictions, and which agencies are responsible for specific decisions. No immediate response was received from the addressed offices or departments.
This congressional inquiry follows similar concerns raised at the state level, where 15 attorneys general had previously sought more information from OpenAI regarding the security incident at Hugging Face.
