US Nears Deal With AI Labs on Voluntary Model Release Standards

A New Framework for Frontier AI

The U.S. government is on the verge of a deal. A big one. Top officials are finalizing a set of voluntary AI safety standards with leading artificial intelligence companies to govern the release of powerful new models. This collaborative framework, which could drop as soon as next week, is all about creating a predictable, secure process for deploying cutting-edge AI. It's a badly needed replacement for the chaotic, ad-hoc approach that has recently roiled the industry.

But make no mistake: this isn't about consumer protection. The real driver here is the grave national security risk—the terrifying possibility that adversaries like China or Russia could weaponize advanced AI. That's the fear. And it's why top-tier labs, including OpenAI, Google DeepMind, and Anthropic, are deep in these talks with the White House.

This move just formalizes what's already been happening behind the scenes. The government has recently been intervening directly in major releases, asking OpenAI to limit its GPT-5.6 launch to a small pool of vetted users and even slapping—then lifting—temporary export controls on Anthropic's new Fable and Mythos models over cybersecurity fears. These new standards aim to bake those reviews into the development lifecycle itself, giving developers some much-needed stability after what can only be described as frontier AI's wild June.

From Executive Order to Handshake Agreement

So where did this all start? President Donald Trump's June 2, 2026, executive order, titled “Promoting Advanced Artificial Intelligence Innovation and Security.” That order kicked things off, directing federal agencies to work with developers on a pre-release framework. It deliberately stopped short of a mandatory licensing regime, something the administration has explicitly rejected. The idea was to find a balance between pushing innovation and protecting the nation's critical infrastructure from new AI threats.

Under that order, agencies like the National Security Agency and NIST's Center for AI Standards and Innovation got their marching orders: build a classified process to benchmark advanced models. This would help figure out when a system is powerful enough to be labeled a “covered frontier model,” which in turn triggers a much heavier review. The new voluntary standards are the direct result of that directive. Policy becomes practice.

The framework contemplates a government review period of up to 30 days before a model is released to other "trusted partners." That's a significant change, representing a much more active federal stance on AI-driven cyber risk and forcing frontier AI companies to build a whole new step into their deployment timelines.

The Global Stakes and the 'Voluntary' Question

While the agreement is being sold as voluntary, the government holds a pretty big stick. Its recent willingness to deploy export controls means any company that opts out of the pre-release review faces a serious business risk. It's a pivot toward a national security-first approach to AI governance, designed to choke off access to the most powerful American-made models for any foreign businesses, governments, or people considered a threat.

These discussions aren't just about who gets access inside the U.S. They're also about who can use these tools overseas, potentially setting a standard for America's allies. This cooperative, industry-led approach is a world away from the European Union's rigid, legally binding AI Act. The U.S. is betting on flexibility over heavy-handed rules. It's a debate playing out globally, with organizations like the UN warning that without coordination, the AI governance gap could deepen the world's digital divide, a key finding in a recent UN report.

But the core question remains. Is a voluntary system really enough? Critics argue that self-regulation is a flimsy defense against the risks of increasingly autonomous, powerful AI. Washington is still debating whether these handshakes should eventually become law. And other federal bodies are already taking a harder line—the FTC, for one, has been quite vocal about prosecuting harmful AI, part of a broader regulatory pressure campaign against things like 'deceptive AI'.

For now, the White House and the AI labs are betting on collaboration. They're hoping a nimble, adaptive system can ensure safety without strangling the very innovation that made the U.S. the world's AI leader. With an announcement imminent, the tech world is holding its breath to see the details of this new pact—and if it will be enough to manage the peril and the promise of frontier AI.