US to Unveil Voluntary AI Guardrails Next Week, FT Reports

The United States is finalizing voluntary AI guardrails for the most advanced AI models, with an announcement possible as soon as next week, the Financial Times reported on July 1, citing people familiar with the talks. The framework would set common benchmarks, deployment timelines, and rules for who can access frontier AI models inside the United States and abroad. The talks are between the Trump administration and leading developers including OpenAI, Anthropic, and Google.

The headline reads like a routine policy story. The shape underneath is more complicated. The US government has, in the past month, intervened directly in the launches of two of the most powerful AI systems of 2026: Anthropic’s Fable 5 and Mythos 5, and OpenAI’s GPT-5.6. The “voluntary” framework now taking shape is, in effect, a written version of what those interventions already look like in practice.

What the FT Report Says Is Coming

The FT reported on Wednesday July 1 that discussions between the US government and AI developers have reached an advanced stage. The proposed standards would cover performance benchmarks, deployment schedules, and access rules for frontier AI models in the United States and overseas, the FT said, citing people familiar with the matter. The framework is expected to be non-binding, the report said.

The White House, Anthropic, and OpenAI did not immediately respond to requests for comment outside regular business hours, Reuters reported. Google has separately been in talks with US authorities ahead of the release of advanced coding-focused models, according to a source familiar with the matter. The same source said Google is also part of the broader industry-standards conversation. The FT first reported the framework details, and Reuters could not immediately verify them.

The June Order That Set the Stage

President Donald Trump signed Executive Order 14409, “Promoting Advanced Artificial Intelligence Innovation and Security,” on June 2, 2026. The order directs federal agencies to design a voluntary framework for AI developers to engage with the federal government before releasing covered frontier models. Agencies are required to finalize that framework within 60 days, by August 1. The order relies on voluntary cooperation from the leading AI developers, including OpenAI, Anthropic, and Google. The full text of Executive Order 14409 was published the same day, alongside a White House fact sheet on the order.

The order asks AI companies to voluntarily submit their most powerful models for the government to test up to up to 30 days before releasing them to “trusted partners.” An earlier draft gave the government up to 90 days, NPR reported, citing people familiar with the deliberations; that window was cut to 30 days in the final order over concerns that a longer review would interfere with American competitiveness against China. The order does not impose licensing, preclearance, or permitting requirements on AI developers.

Section 3 of the order sets the playbook. The Secretary of the Treasury, the Secretary of War through the Director of the National Security Agency, and the Secretary of Homeland Security through the Director of the Cybersecurity and Infrastructure Security Agency are directed to develop a classified benchmarking process. That process will determine which AI models are designated “covered frontier models.” The Director of the NSA will make those designations in consultation with the National Cyber Director, the White House Office of Science and Technology Policy, and CISA.

Government Is Already Shaping AI Launches

The voluntary framework is being pre-written by specific interventions that have already changed two of 2026’s biggest model launches. Anthropic’s earlier announcement in April that it was limiting the release of its Mythos Preview model, citing the model’s ability to find software vulnerabilities, set off alarm bells across Silicon Valley and Washington, NPR reported. The Department of Commerce sent Anthropic a letter on June 12 imposing export controls on its newest models, Fable 5 and Mythos 5, after Amazon researchers reportedly identified ways to bypass the guardrails on Fable 5. Anthropic disabled access to both models for every customer, including its own foreign-national employees, in order to comply. The order cited “national security authorities,” according to a CSIS analysis of the underlying authority. The letter has not been made public, but media reports and people who have seen it described its scope to CSIS.

The Commerce Department lifted the controls on June 30, less than three weeks after the original order. Anthropic said in a post on X that “all export controls on the models have now been lifted after the implementation of safeguards.” US Commerce Secretary Howard Lutnick had earlier given Anthropic permission to release Mythos 5 to a select group of companies and federal agencies on June 26. The negotiations were led on Anthropic’s side by co-founder Tom Brown, CNBC reported, replacing Chief Executive Dario Amodei in the role. A CSIS analysis of the export-control mechanism explains the Bureau of Industry and Security’s “is informed” authority and why the case now sets precedent.

OpenAI was asked by the Trump administration the same week to limit the launch of GPT-5.6 to a small group of vetted partners. It was the first time the US government had preemptively asked an American AI company to restrict the launch of a model before release, according to Axios. The reason, a source told Axios, was that GPT-5.6 has “Mythos-like” capability and the administration wanted assurance that adequate safeguards were in place.

For Anthropic’s Fable 5 and Mythos 5 specifically, the result was a new industry jailbreak-severity framework, alongside the lift on June 30. Anthropic’s Fable 5 returning to global users under new rules walks through the safeguard structure the company agreed to. Companies that build models with cyber-capable frontier capabilities should plan around a possible 30-day window before any release to “trusted partners.” Customers waiting for new frontier models should expect similar delays.

Who Is in the Room

Three companies lead the formal talks: OpenAI, Anthropic, and Google. OpenAI delayed GPT-5.6’s broader release on June 27 after the Trump administration requested early access and approved-customer review. In its announcement, OpenAI said it had previewed the model’s capabilities and shared its plans with the government ahead of the launch.

Anthropic is the most exposed company in the room. Fable 5 and Mythos 5 were the catalysts for both the executive order and the new framework; the company’s Tom Brown led negotiations in Washington, while Chief Executive Dario Amodei stayed out of the talks. Google is in separate talks ahead of advanced coding-focused models, the FT reported, citing people familiar with the matter. The talks with Google are part of the broader industry-standards conversation, a source told Reuters.

The IPO Clock and Why Timing Matters

Both OpenAI and Anthropic are preparing for initial public offerings. Anthropic confidentially filed for a US IPO in early June, Reuters reported, weeks after raising $65 billion at a $900 billion post-money valuation. OpenAI’s advisers have presented the option of waiting until 2027 to go public with a $1 trillion valuation, the New York Times reported.

A clear rulebook for releasing frontier models is now part of the pitch to public-market investors. Voluntary pre-release engagement that becomes standard could shorten the runway between model announcement and full release. It could also let competitors, regulators, and adversaries read deployment timelines before models reach the public. The August 1 framework deadline will be one of the first policy signals that public investors will weigh. The $852 billion valuation OpenAI fetched in early 2026 is already under scrutiny from its own backers as the company shifts focus to enterprise markets, the FT has reported. Public-market investors will price the framework’s effect on deployment schedules into any IPO multiple.

OpenAI has pushed back on the direction of travel. “We don’t believe this kind of government access process should become the long-term default,” the company wrote in its Friday announcement. CEO Sam Altman told employees in a memo that the company had made clear to the US government that this is “not our preferred long term model.” Altman said he hoped to release GPT-5.6 a “couple of weeks later,” according to The Information.

• 30 days: maximum pre-release government access under the June 2 order
• 60 days: deadline for agencies to finalize the voluntary framework (by August 1, 2026)
• $900 billion: Anthropic’s post-money valuation after its late-May raise
• $1 trillion: OpenAI valuation advisers have floated for a possible 2027 IPO

What “Voluntary” Actually Means

The order is explicit about what it does not do. Section 3 says “nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models.” That language appears in the White House fact sheet accompanying the order. It is the part of the framework the administration has chosen to repeat most often. A legal analysis of the voluntary framework explains how the disclaimer interacts with the classified benchmarking process.

In practice, the Commerce Department already used a separate, non-voluntary lever to restrict Anthropic: a letter imposing license requirements under the Bureau of Industry and Security “is informed” authority, a mechanism that had never been used before as the basis for an export control, according to the Center for Strategic and International Studies. The same mechanism is what forced Fable 5 and Mythos 5 offline for 18 days. That non-voluntary precedent now sits inside a “voluntary” framework.

Frequently Asked Questions

When will the US AI guardrails be unveiled?

The Financial Times reported on July 1, 2026 that an announcement is possible as soon as next week. The June 2 executive order gives agencies until August 1, 2026 to finalize the framework.

Which AI companies are part of the talks?

The Financial Times and Reuters named OpenAI, Anthropic, and Google as participants. Google has been in separate talks ahead of advanced coding-focused models.

What is a covered frontier model?

It is a designation the Director of the NSA will assign under a classified benchmarking process. The order directs agencies to set the threshold within 60 days of June 2, 2026.

What did the June 2 executive order actually do?

Executive Order 14409 directs federal agencies to design a voluntary framework for AI developers to engage with the federal government before releasing their most advanced models. The order expressly disclaims any licensing or preclearance requirement.