The landscape of artificial intelligence regulation in the United States is on the cusp of a potentially transformative shift. According to a report from the Financial Times, the Trump administration is finalizing a voluntary standards deal with several of the country's leading frontier AI companies. The agreement, which could be announced as early as next week, is expected to focus heavily on cybersecurity capabilities and the safe deployment of advanced AI models.
The report, citing anonymous sources familiar with the negotiations, indicates that the deal will be spearheaded by two key government bodies: the Center for AI Standards and Innovation (CAISI), which operates under the Commerce Department, and the National Security Agency (NSA), part of the Pentagon. These agencies will be central to formalizing the benchmarks that AI developers must meet to avoid regulatory crackdowns.
This development comes against a backdrop of increasing tension between the White House and AI companies. In mid-June, the U.S. government issued an export control directive to Anthropic that effectively shut down the company's latest publicly released model for the remainder of the month. OpenAI, reportedly fearing a similar disruption, has held back its newest models as a precautionary measure. These actions represent a stark departure from the early days of the Trump 2.0 administration, when Vice President J.D. Vance signaled a laissez-faire approach to AI oversight.
The proposed standards are expected to establish a classified benchmarking process to evaluate the advanced cyber capabilities of AI models. Under a previous executive order issued by President Trump, the government is tasked with developing and maintaining this process to determine the threshold at which an AI model should be designated a 'covered frontier model.' The goal is to share these assessments with AI developers and researchers as appropriate—though the classified nature of the benchmarks means the public will not have direct visibility into the specific criteria.
The companies reportedly involved in the voluntary vetting agreement include Anthropic, OpenAI, Amazon, Microsoft, and Google. Notably absent from the list is Meta, which last week was reported to be a holdout in the negotiations. The Trump administration has been working overtime to secure Meta's buy-in, though the reasons for the company's reluctance remain unclear. Some analysts speculate that Meta's open-source approach to AI development, such as its Llama models, may clash with the government's desire for stricter controls.
The push for voluntary standards is a significant step in the broader narrative of AI regulation. For years, the industry has operated in a gray area, with companies racing to deploy increasingly powerful models while governments around the world scramble to create guardrails. The United States has so far favored a light-touch approach, but recent actions suggest a hardening stance. The new standards, if implemented, could bring much-needed clarity to what AI companies are and are not allowed to do, particularly regarding national security.
Cybersecurity is a particularly urgent concern. Frontier AI models can be used both defensively and offensively; they can help patch vulnerabilities but also create sophisticated malware or automate phishing campaigns. By focusing on cyber capabilities, the administration is addressing a concrete threat that has bipartisan attention. The classified nature of the benchmarks, however, raises transparency concerns. Critics argue that without public oversight, the standards could be used to stifle competition or protect incumbent firms.
The deal also has implications for international competition. Europe is already moving ahead with its AI Act, which imposes strict requirements on high-risk AI systems. China, meanwhile, has implemented its own set of regulations. A U.S.-backed voluntary framework could serve as a template for other nations, positioning American companies as responsible leaders—or as cozily intertwined with government priorities, depending on one's perspective.
For the companies involved, the agreement offers a measure of predictability. Instead of facing ad hoc crackdowns like the one against Anthropic, they would have clear standards to meet. But the voluntary nature of the deal means there is no legal obligation to comply—at least for now. Some lawmakers and advocacy groups have called for mandatory rules, arguing that voluntary commitments have failed to prevent AI harms in areas like bias, misinformation, and privacy.
As the announcement approaches, all eyes are on whether Meta will join the pact. The company's absence could signal a rift in the industry, potentially leading to a two-tier system where compliant firms benefit from government favor while holdouts face scrutiny. It also raises questions about the effectiveness of standards that lack universal adoption.
In the meantime, the public remains largely in the dark about the specifics of the benchmarks. The executives of OpenAI, Google DeepMind, and other firms have been tight-lipped, while administration officials emphasize national security considerations. What is clear is that the Trump administration has moved from a hands-off posture to one of active engagement, setting the stage for a new era of AI governance in the United States.
The roots of this shift can be traced back to the rapid advancement of models like GPT-4 and Gemini, which have surpassed previous capabilities at an unprecedented pace. With each new release, the potential for misuse grows, prompting governments to intervene. The voluntary deal represents a middle path—an attempt to impose order without the heavy hand of legislation. Whether it will be sufficient remains to be seen.
Industry observers note that the deal is likely to include provisions for regular testing and reporting, as well as mechanisms for independent audits. The NSA's involvement suggests a focus on preventing AI from being used to launch cyberattacks on critical infrastructure. CAISI, on the other hand, is expected to develop technical standards that can be adopted across the industry, similar to how safety standards are set for aviation or pharmaceuticals.
One of the most contentious issues is the handling of open-source models. Meta's Llama series is widely used by developers and researchers, but its openness also makes it harder to control. The government's approach to open-source AI could have far-reaching consequences for innovation. If the standards effectively require close oversight of model releases, smaller companies and academic labs may struggle to compete.
The timing of the announcement—coming in early July 2026—also has political implications. With midterm elections approaching, the administration likely wants to showcase a proactive stance on AI safety, a topic that increasingly resonates with voters concerned about job displacement, deepfakes, and national security. At the same time, overregulation could be portrayed as stifling American innovation, a charge the White House is eager to avoid.
For now, the details remain under wraps. The Financial Times report provides the clearest picture yet of what is to come, but many questions persist. Will the standards be made public in any form? How will compliance be verified? And what happens if a company breaches the voluntary agreement? These are issues that will play out in the coming weeks and months.
What is certain is that the relationship between the U.S. government and Big AI is evolving rapidly. From a period of relative non-intervention, we are moving toward a regime of structured cooperation—or control, depending on one's viewpoint. The next few days will reveal whether Meta comes on board, and whether the standards can actually deliver the safety and predictability that policymakers and the public are demanding.
Source:Gizmodo News