The Trump administration is accelerating the development of a frontier AI regulatory framework, requiring major firms such as OpenAI, Anthropic, and Google to undergo up to 30 days of government review before publicly releasing their most advanced models.
Central points of contention in the framework include the boundary defining 'frontier models' and whether open-source models should be exempt. Doubts also surround the credibility of 'voluntary compliance' commitments—recent export controls imposed by the White House on Anthropic have led the industry to question the meaning of 'voluntary.'
The Trump administration is accelerating efforts to advance a regulatory framework targeting cutting-edge AI models, requiring companies to submit their most advanced models for government review prior to public release. A draft of the framework was circulated to major tech giants several days before the deadline.
On Tuesday, according to The Information citing three sources familiar with the matter, the White House Office of the National Cyber Director distributed a draft of the framework to OpenAI, Anthropic, and$Alphabet-C (GOOG.US)$, after which the three companies jointly submitted proposed revisions. The framework stems from an executive order signed by Trump in early June, setting an August 1 deadline for completion. It aims to establish a more structured process for government review of powerful AI models, with particular focus on frontier models capable of rapidly identifying cybersecurity vulnerabilities.
This move will provide clearer regulatory grounding in what has previously been an uncertain AI oversight environment. Since the executive order was issued, the White House has imposed export controls on Anthropic and required OpenAI to roll out its latest model, GPT-5.6, in stages. These actions have drawn criticism from the industry, which argues that the government is effectively imposing an ad hoc licensing regime on the AI sector. The release of this framework is seen as a key initiative by the White House to bolster U.S. AI development amid intense competition with China.
Origins of the Framework: Executive Order Addressing Safety Risks of Frontier Models
The immediate catalyst for the executive order was Anthropic’s frontier model, Mythos. Due to the model’s offensive cybersecurity capabilities, Anthropic proactively delayed its large-scale public release. In seeking a response, the Trump administration aimed to maintain its generally hands-off stance on AI regulation while ensuring effective oversight of national security risks posed by new models.
According to informed sources, the final version of the executive order underwent multiple rounds of negotiation. David Sacks, former head of AI and cryptocurrency affairs, intervened at the last minute, blocking Trump from signing the initial draft on grounds that its provisions were overly burdensome for businesses. Subsequently, Trump signed a revised version on June 2, stipulating that AI labs must grant the federal government up to a 30-day access window before releasing any 'covered frontier model' to other partners, and giving government agencies 60 days to design a detailed framework.
Since June, the White House has held multiple feedback sessions with companies, including a joint meeting on June 9, to finalize the framework ahead of the August 1 deadline.
Core Dispute: Defining Frontier Models
One key area of disagreement concerns how to define the scope of 'frontier models.' According to two informed sources, some smaller AI firms worry that the definition was primarily benchmarked against OpenAI, Anthropic, and$Alphabet-C (GOOG.US)$the three largest labs, potentially excluding them from coverage.
Criteria for identifying frontier models span multiple dimensions, including model scale, deployment method, and specific capabilities assessed by third-party evaluations. If the definition focuses solely on leading closed-source models such as Mythos and GPT-5.6, it could exclude open-source models—typically trained with fewer parameters and often deployed locally rather than in the cloud—even if their capabilities are comparable to those of closed-source counterparts.
For companies included in the framework, participation in the program carries potential benefits: the executive order will direct additional funding toward critical infrastructure and cybersecurity defenses, although the actual scale and allocation mechanisms of these resources remain unclear.
Open-Source Models: The Debate Over Regulatory Exemptions
How open-source models are treated represents another unresolved core issue. According to prior reporting by The Information, parallel discussions have already taken place between companies and the government regarding whether open-source models should receive voluntary review exemptions based on their capabilities, in order to alleviate additional regulatory burdens.
Proponents of exemptions argue that publicly available open-source models in the United States currently lag behind their Chinese counterparts. Figures such as David Sacks advocate for reduced regulation to ensure room for catching up. If the framework does not classify open-source models as frontier models, they will be excluded from the program and cut off from access to relevant resources.
The Actual Enforceability of the Voluntary Framework Remains Questionable
Although the Trump administration insists that compliance with the framework is not mandatory, its de facto enforceability remains one of the industry’s primary concerns. The White House’s previous decision to impose export controls on Anthropic has raised doubts about the credibility of so-called 'voluntary' commitments.
During a press briefing in May this year, when asked what consequences companies would face if they refused to submit their models, a senior White House official declined to provide a direct answer, stating only that the framework 'is designed to be flexible and evolve alongside emerging technologies' and that it 'strikes an appropriate balance between safety and innovation.'
Regarding the review mechanism, the White House has informed certain companies that model reviews will be jointly conducted by the National Security Agency (NSA) and the Commerce Department’s Center for AI Standards and Innovation (CAISI). According to informed sources, CAISI faced political pressure during the Trump administration, but optimism is now growing externally regarding the White House’s recognition of its technical expertise.
Editor/melody