The White House is in advanced talks with artificial intelligence (AI) companies to create voluntary AI model standards for new frontier releases. The standards, which could be announced as soon as next week, are expected to set benchmarks for models with advanced cyber capabilities and establish release timelines. The move follows President Trump’s June executive order directing agencies to draft standards for AI model testing and deployment.
What the voluntary AI model standards would cover
The proposed framework is designed to create a common set of voluntary AI model standards for evaluating highly capable AI models before their release, replacing the recent case-by-case approach that has left developers uncertain over regulatory expectations.
The discussions cover defining what qualifies as “frontier” AI, how long safety evaluations should take, and what security benchmarks must be met before public deployment.
To this point, officials also plan to clarify who can access the most advanced models in the U.S. and overseas, potentially laying the groundwork for a broader framework among U.S. allies.
Technical teams from OpenAI, Anthropic, and Google have been meeting regularly with government officials to work out the details. It is a pretty big deal that these standards are voluntary in nature, especially since the administration has passed on the idea of creating a brand-new AI regulator. Instead, they are sticking with teams that are already on the job, like the National Institute of Standards and Technology (NIST) and the Cybersecurity and Infrastructure Security Agency (CISA), to keep an eye on things.
Recent interventions fuel the push
The voluntary AI model standards effort follows a string of government interventions involving next-generation AI systems. Earlier this month, the U.S. Commerce Department imposed temporary export restrictions on Anthropic’s Fable and Mythos models over cybersecurity concerns before lifting them this week.
OpenAI was also asked to limit access to its upcoming GPT-5.6 model to government-approved users ahead of a broader release.
These case-by-case interventions, driven by concerns that advanced AI could be misused by military intelligence in China, Russia, or other adversarial nations, have left companies seeking clearer rules. So these new voluntary AI model standards are intended to provide that predictability while preserving American innovation in the global AI race, which is openly growing every day.
