Trump Administration Preparing Voluntary Security Standards for Top AI Firms

The Trump administration is preparing to announce a voluntary agreement with the largest American artificial intelligence companies covering safety standards for their most advanced models. According to reports, the announcement could come as early as this week, though details remain largely under wraps.

What the deal covers

At the core of the arrangement is a classified benchmarking process meant to assess advanced cyber capabilities in AI models and determine which ones qualify as so-called covered frontier models, subject to heightened oversight. Because the evaluation process itself stays classified, the public won't know the exact criteria used to classify the models.

Formalizing the standards would fall to two agencies: the Center for AI Standards and Innovation (CAISI), housed under the Department of Commerce, and the National Security Agency (NSA), tied to the Pentagon. Companies covered by the agreement would be expected to publish their own risk assessment frameworks, disclose their models' capabilities and the safeguards in place to limit risks, and report security incidents to the relevant regulators.

Who is signing, who is hesitating

OpenAI, Anthropic, Google, Microsoft and Amazon were invited to the talks. Meta remains the holdout so far, and the administration reportedly pushed hard in the final days before the planned announcement to bring it on board.

The fact that the US government is negotiating voluntary rules with companies rather than imposing hard statutory regulation fits the Trump administration's broader approach so far, favoring cooperation with industry over rigid rules that could slow down a technology seen as key to US competitiveness against China.

The wider context of talks with Big AI

The agreement arrives amid intense negotiations between Washington and AI industry leaders on multiple fronts at once. OpenAI recently offered the US government a roughly 5 percent stake in the company to ease political pressure over its growing role in the economy and national security. Trump had also previously floated, more broadly, the idea of the government taking direct equity stakes in leading AI companies.

These parallel moves, voluntary safety standards on one hand and government equity stakes on the other, show how hard the administration is trying to both rein in the risks posed by the most powerful models and capture the upside from their commercial success.

What it means outside the US

For companies and institutions using OpenAI, Anthropic or Google models outside the United States, including in Poland, the new standards could in practice mean additional delays in the rollout of the most advanced models if they have to clear a pre-release verification process first. The secrecy around the evaluation criteria will also make it harder for third-party firms to predict which model features might get restricted for safety reasons.

The final shape of the agreement, including whether Meta ultimately joins, should become clear in the coming days. It would mark the Trump administration's first concrete step toward regulating frontier model safety, after months of announcements and informal talks with the industry.

Sources: Trump Administration Reportedly on Verge of Standards Deal With Big AI (gizmodo.com)