US in talks with AI companies on voluntary standards for new model releases

The US government is negotiating with leading AI companies over voluntary standards for releasing new models. According to the Financial Times, an official announcement could come soon.

What the two sides are negotiating

According to the Financial Times, the standards under discussion would set benchmarks and timelines for advanced AI models and clarify who gets access to those models, both inside the US and abroad.

• The talks are between the US government and AI companies
• The standards are voluntary, not legally binding
• The standards cover the process for releasing new models
• Access to models at home and abroad is on the table
• The source of the information about the talks is the Financial Times

The fact that these talks are happening at all suggests the federal government is looking for an alternative both to a fully hands-off approach to AI development and to the kind of strict legislative regulation some individual states have pursued. The Financial Times does not name the companies involved, but given the scope of the issues on the table, from benchmarks to controlling access abroad, the talks clearly involve the largest developers of advanced models.

Why a voluntary format

Voluntary commitments let the government and AI companies agree on the rules of the game without going through a lengthy legislative process. For companies, that means predictability: benchmarks and timelines known in advance instead of a patchwork of requirements that can emerge at the state level.

For AI labs, voluntary standards reduce the risk of being subject to several parallel, and sometimes contradictory, requirements from multiple states at once. At the same time, because the agreement is voluntary, federal authorities may have no direct enforcement levers at this stage: companies either sign on to the standards or stay outside the framework.

This also fits the broader federal approach to tech regulation in recent years. Washington has consistently preferred targeted agreements with industry over centralized lawmaking, arguing that rigid federal rules simply would not keep pace with how fast the technology itself is evolving.

What the access rules mean

A separate part of the talks concerns who gets access to new advanced models, both inside the US and beyond its borders. This ties directly to national security and to controlling the spread of the most powerful AI systems among foreign players, including potential geopolitical rivals.

Framed this way, the standards turn the talks into more than a product safety question: they become a foreign policy tool. Access to frontier models becomes a resource the US government can regulate alongside exports of technology and hardware. For US allies, this could mean more predictable access to frontier models, while for countries under export restrictions, the talks continue the current course of technological protectionism.

What this means

If the talks end in an official announcement, the US will get its first notable benchmark for industry self-regulation: a compromise between the desire to control the release of increasingly powerful models and the reluctance to slow the competitive race with strict legislation. For the rest of the world, it will signal which path the United States ultimately chose: soft law instead of direct regulation. Until the official announcement, the details remain under negotiation, which means the final set of requirements may differ from what Washington and the largest AI labs are currently discussing behind closed doors.

The Financial Times says it will keep following the talks and update its reporting as new details about the content of the future standards emerge.