The Trump administration is reportedly on the verge of announcing a voluntary standards deal with several of the largest frontier AI companies, marking a significant shift in the federal government's approach to artificial intelligence regulation. According to the Financial Times, which cited anonymous sources familiar with the matter, the agreement could be unveiled as early as next week and will primarily address cybersecurity capabilities of advanced AI models.
The reported standards deal would involve the Center for AI Standards and Innovation (CAISI), a body housed under the Department of Commerce, and the National Security Agency (NSA), which falls under the Pentagon. These agencies are expected to play a central role in formalizing and overseeing the new requirements. The move comes after a period of relative regulatory uncertainty that has left the AI industry navigating a patchwork of expectations and potential enforcement actions.
The development represents a notable departure from the initial posture of the Trump 2.0 administration. At the start of the term, Vice President J.D. Vance signaled a largely laissez-faire approach to AI regulation, suggesting the government would refrain from heavy-handed interference in the technology's development. However, recent events have demonstrated a more interventionist stance. In June, the U.S. government issued an export control directive to Anthropic, effectively forcing the company to take its latest publicly released model offline for the remainder of the month. This action sent shockwaves through the industry, leading OpenAI—another leading AI firm—to withhold the release of its newest models as a precautionary measure, fearing similar treatment.
The administration's shift was further codified in a recent executive order on artificial intelligence, which was described as a “national security” priority. The order mandates the creation of a classified benchmarking process to assess the advanced cyber capabilities of AI models. It also establishes a threshold for designating a model as a “covered frontier model,” a label that would trigger additional oversight and requirements. The classified nature of the benchmarking process has raised concerns among transparency advocates, who argue that the public will have little insight into the standards to which the most powerful AI systems are being held.
According to the FT report, the companies expected to participate in this voluntary vetting agreement include Anthropic, OpenAI, Amazon, Microsoft, and Google. Notably absent from the list is Meta, the parent company of Facebook and Instagram. Around a week ago, other anonymous sources familiar with the negotiations leaked that Meta has been a holdout in the talks, and that the Trump administration has been working “overtime” to secure the company's buy-in. Meta's reluctance may stem from its open-source approach to AI development, which often conflicts with the kind of centralized, classified oversight that the deal envisions.
The proposed standards primarily focus on cybersecurity capabilities of frontier AI models. The government, through CAISI and the NSA, intends to define and assess how these models can be used—or abused—for cyberattacks. This includes evaluating their ability to autonomously discover vulnerabilities, craft sophisticated phishing campaigns, or even develop novel malware. By establishing a shared set of practices and safeguards across multiple companies, the administration hopes to create a baseline for national security without imposing overt regulations that would require congressional approval.
Critics, however, warn that the voluntary nature of the deal may lack teeth. Without mandatory compliance, companies could choose to ignore the standards if they become competitively disadvantageous. Moreover, the classified nature of the benchmarks means that companies themselves may not fully understand the criteria by which their models are judged, potentially leading to confusion or mistrust. On the other hand, proponents argue that a voluntary agreement allows for more flexibility and faster adaptation to the rapidly evolving AI landscape, avoiding the lengthy and often outdated process of formal legislation.
The backdrop to this deal includes a broader global conversation about AI safety and regulation. The European Union has already passed the AI Act, which imposes risk-based obligations on AI developers. China, meanwhile, has implemented strict controls on AI exports and algorithms. The United States has yet to pass comprehensive federal AI legislation, leaving agencies like the Commerce Department and the NSA to fill the gap through executive orders and voluntary agreements.
The involvement of the NSA is particularly noteworthy. The agency has long been at the forefront of both offensive and defensive cyber operations. Leveraging its expertise in cryptography and vulnerability assessment, the NSA is well-positioned to evaluate the cyber capabilities of frontier AI models. However, its history of surveillance and secrecy may fuel concerns about the extent of its oversight role. Some civil liberties groups have already expressed alarm that the NSA could use the benchmarking process to expand its domestic monitoring powers, while national security hawks counter that such oversight is essential to prevent AI-powered cyberattacks from rogue states or terrorists.
The industry response has been mixed. Anthropic and OpenAI, which have faced direct government actions in recent weeks, are likely eager to establish clear rules of the road to avoid future surprises. Microsoft and Google, with their extensive government contracting arms, are accustomed to navigating federal security requirements. For Amazon, the deal could help clarify how its AWS cloud services and AI offerings meet national security standards. The holdout by Meta suggests that the company may be willing to risk government ire in favor of maintaining its open-source ethos, which it argues accelerates innovation and democratizes access to AI.
The timing of the deal is also significant. It comes as the Biden-era executive order on AI is being partially rescinded or rewritten, and as the Trump administration seeks to differentiate its approach. Whereas the previous administration emphasized fairness, civil rights, and transparency, the Trump team is framing AI regulation primarily through the lens of national security and economic competitiveness. The new standards deal reflects that prioritization, focusing squarely on cybersecurity capabilities rather than issues such as bias or job displacement.
Looking ahead, the announcement of the voluntary standards could have far-reaching implications. It may serve as a template for other nations seeking to manage the dual-use nature of frontier AI. It could also set a precedent for future agreements involving other sensitive domains, such as biosecurity or autonomous weapons. However, the absence of a transparent, public-facing standard may erode trust among users and advocates who fear that AI companies and the government are making secret bargains that prioritize security over accountability. The tension between secrecy and openness will likely remain a central theme as the details of the deal are finalized.
In the coming days, more details are expected to emerge from leaks and official statements. For now, the industry and the public can only wait to see whether the voluntary standards will truly provide clarity and safety, or whether they represent another chapter in the complex dance between innovation and control.
Source: Gizmodo News