In partnership with
Washington Wants Frontier AI Standards Without Frontier AI Laws. Here Is What That Looks Like.
In early July, the Financial Times reported that the U.S. government is in advanced talks with leading AI companies to set voluntary standards for releasing new models, with an announcement possible as soon as next week. The discussions involve OpenAI, Google, Anthropic and others and focus on the models at the very top end of capability.
The standards under negotiation would set benchmarks for advanced models, define timelines for review and clarify who can access those models in the United States and abroad. The structure is deliberate. It is designed to shape behaviour without turning into a licensing regime.
Porkbun is the domain name registrar you need.
Still using GoDaddy or Namecheap? There’s a better way with Porkbun!
Porkbun is the domain registrar trusted by creators, developers, entrepreneurs, and folks who want low prices without the nonsense.
Why people are choosing Porkbun:
• Most domains sold at cost
• Low, transparent registration and renewal pricing
• Free features like WHOIS privacy and SSL certificates
• Powerful web and email hosting options
• Real human support 24/7, 365 days a year
• Named the #1 domain registrar by Forbes Advisor and USA Today
For launching a business, building a personal brand, starting a side project, or creating your first website, Porkbun makes it easy.
Get $1 off your next domain registration with Porkbun now.
The Policy Architecture Behind These Talks
These negotiations are not happening in a vacuum. They sit on top of an executive order that the White House issued in early June.
President Trump’s order on “Promoting Advanced Artificial Intelligence Innovation and Security” directs agencies to develop a classified benchmarking process for advanced cyber capabilities in AI models and to determine when a system should be designated as a “covered frontier model”. That designation is central. It triggers the framework now being negotiated.
The order instructs the NSA, CISA, Treasury and other agencies to maintain a confidential benchmark against which model capabilities can be measured. If a model crosses the threshold, it becomes eligible for inclusion in a voluntary early access scheme.
The same order then calls for a voluntary framework with developers. Under that framework, companies can:
Engage the government to determine whether a model under development is a covered frontier model.
Provide the government secure access to such models for up to thirty days before broader release.
Collaborate with the government on selecting trusted partners who will receive early access.
Crucially, the text explicitly states that nothing in the order authorises mandatory licensing, preclearance or permitting for the development or release of models. The legal line is drawn very clearly.
The talks now underway about voluntary standards are essentially the practical translation of that order into day‑to‑day rules.
What the Voluntary Standards Are Likely to Cover
Across Reuters, FT and other summaries, two themes show up repeatedly: benchmarks and timelines, and access rules.
Benchmarks and thresholds. The standards will define capability benchmarks that distinguish ordinary models from frontier ones. These are likely to build on the classified benchmarking process described in the June order, especially around cyber capabilities and potential misuse in sensitive contexts.
Timelines for review. The order already mentions a thirty day window in which agencies can test covered frontier models before companies release them to other trusted partners. Negotiations now appear to be focused on how long that window can be in practice and whether it can vary depending on the risk profile of a given system.
Access rules. The FT and Reuters report that the standards will clarify who can access frontier models inside the United States and abroad. That likely includes guidance on export restrictions, geographic availability and which kinds of organisations are considered “trusted partners” during the early access phase.
Google is reportedly in discussions with the government ahead of the release of its own advanced coding models, which are expected to have more sophisticated cyber capabilities than prior generations. That is a concrete example of where these standards will apply first.
Why “Voluntary” Is Doing So Much Work Here
The voluntary framing is doing two jobs at once.
On one side, it reassures developers that the government is not trying to create a licensing regime. The executive order goes out of its way to say that nothing in it should be read as authorising mandatory permits for model development or release. In political terms, this avoids the fight that a more intrusive framework would immediately provoke.
On the other side, it still gives Washington a structured way to see, test and influence the deployment of the most capable systems. Agencies are being asked to secure agreements with developers to submit frontier models voluntarily for cybersecurity assessment before release, with a thirty day testing window. The current talks about standards are an extension of those agreements.
For large U.S. based developers, “voluntary” in this context is not the same as optional in practice. These companies operate in a regulatory environment where ignoring a national security request can carry reputational and legislative risks. The executive order itself represents a shift from the administration’s earlier reluctance to intervene in technology policy.
In that sense, the voluntary standards are a compromise instrument. They give the government a lever without creating a new legal burden. They give companies some predictability without putting them under formal licensing.
How This Affects Model Release Strategy
For developers working at or near the frontier, the emerging standards introduce a new variable into release planning.
They will need to consider:
Whether a given model crosses the frontier threshold under the benchmarks being defined.
How a thirty day review window affects timelines for public announcements, customer launches and infrastructure planning.
Which customers are considered trusted partners for early access, and how that affects commercial relationships.
If a company like OpenAI or Anthropic accepts the standards, it effectively commits to a pattern where certain classes of models are seen by government and a small set of partners before the wider market. Gemini 3.5 Pro, which is on track for general availability without similar restrictions, may become a useful contrast point for how different developers choose to interact with the framework.
From a user’s perspective, none of this guarantees slower access across the board. It does mean that the release of the most capable systems may be increasingly shaped by an interplay between technical readiness and state comfort, rather than by product planning alone.
What to Watch in the Coming Weeks
If the Financial Times reporting is correct, an announcement could arrive as soon as next week.
Three details will be worth watching closely:
The threshold definition. How the standards define “frontier” in practical capability terms will determine which models fall under this framework and which do not.
The review window. Whether the thirty day period in the executive order becomes the default or whether shorter or longer windows are introduced for particular risk classes, will affect how quickly frontier models reach general users.
The access guidance. Any explicit rules on which sectors, geographies or organisation types can access frontier models will signal how seriously Washington is taking export control and critical infrastructure protection in this context.
For now, the key point is simple. The U.S. is choosing a path where it can shape frontier model releases through standards and agreements rather than laws and licences. That choice keeps the formal regulatory temperature lower but it still moves the relationship between developers and the state into a more closely managed space.