Getty Images The Claude Fable logo is displayed on the screen of a smartphone placed on a laptop keyboard, with the company’s branding illustration on the screen, in Creteil, France, on June 10, 2026. Anthropic announces the release of two new Mythos-class artificial intelligence models for cybersecurity and biomedical research, targeting both consumers and businesses. (Photo by Samuel Boivin/NurPhoto via Getty Images) Earlier this month, Anthropic launched Claude Fable 5, an AI model the company built after cooperating with the federal government in good faith. Anthropic had allowed government reviewers access to its underlying model of Mythos, took their feedback seriously, and added guardrails against high-risk uses in cybersecurity and biological materials. Within days of Fable’s public release, the Department of Commerce hit it with export control restrictions, reportedly after a phone call from Amazon CEO Andy Jassy to the White House.
Anthropic did everything right. It got punished anyway.
Days before the launch of Fable, the Trump administration issued its Executive Order encouraging artificial intelligence companies to voluntarily give the federal government up to 30 days of access to their “covered frontier models,” to promote security and protect critical infrastructure. But “covered frontier model” remains undefined.
The order gives a litany of departments 60 days to set the threshold, including the secretary of the Treasury; the secretary of Defense (via the National Security Agency); the secretary of Homeland Security (via the Cybersecurity and Infrastructure Security Agency, currently without a director); the assistant to the president for Science and Technology Policy; and the secretary of Commerce (via the National Institute of Standards and Technology). Before this process can evaluate a single model, it first has to be built. This will take time.
Other than the scheduling issues such a deadline imposes, the learning curve for review is too high. The current executive order directs that the Treasury secretary lead the process despite having little institutional knowledge in this area. The National Security and Cybersecurity and Infrastructure Security agencies have previously led AI policy coordination, and the Center for AI Standards and Innovation ultimately lead the evaluations of any models. These players need to be integrated and collaborating from the start.
Even if that process comes together, 30 days is not enough time to review new models. AI models are probabilistic and change with use; a model that shows no capacity for harm during a month of government testing may develop one later. Furthermore, government reviewers are going to be hampered by staffing and expertise. Even if the entire AI force of the federal government evaluates the covered frontier model, that expertise may not match the inherent expertise in companies projected to be trillion-dollar firms. A 30-day review by a smaller, less specialized team is not a safety guarantee.
I have seen this trap before. When I was assistant secretary at the Department of Homeland Security for Countering Weapons of Mass Destruction in 2024, frontier model companies came to us asking for exactly this kind of sign-off — confirmation that their models could not help someone build a biological or chemical weapon. We said no, for the same reasons that make this executive order unworkable: the time required, the limited bandwidth, and the fact that AI models change after review.
This month’s order revives that same flawed logic. And it adds a new danger: If a model that passed government review later enables harm, the question of why will not be directed at the company. It will be directed at the government.
The Fable story shows why no company will trust this process. Anthropic cooperated before the executive order required it. The government reviewed the model. Anthropic built in the safeguards. The model launched. Then Commerce restricted it — not because review found a problem, but purportedly because a private competitor’s CEO made a phone call to the White House.
Why would any company submit to a voluntary review process that offers no legal protection and leaves them exposed to exactly this kind of arbitrary reversal?
In fact, submitting to such a procedure may draw more scrutiny to a new model than it would have received otherwise. And as Fable demonstrates, the government may reach for other tools like export controls regardless of what any review may find. A voluntary process with no binding effect and no reciprocal obligation from the government is not a framework. It is a trap.
There is a better model.
The Frontier Model Forum already runs an information-sharing process between AI developers on risks and vulnerabilities, including in cybersecurity and biological materials. Leveraging that process will allow the government to build on top of private-sector information sharing, creating standing joint working groups and understanding long-term increase in risks.
The government and frontier model companies should be meeting continuously, not reviewing models once before launch and walking away. Continuous engagement tracks how models evolve. It builds the expertise on the government side that a 30-day snapshot never could. And it creates a relationship that makes companies want to cooperate, rather than fear the consequences of doing so.
The lesson of Fable is not that cooperation between government and AI companies failed. It is that the government was not ready to hold up its end of the bargain. Until that changes, no serious company will volunteer for this process. And the U.S. will have traded the trust it needs in AI leadership for a 30-day review that protects no one.
Mary Ellen Callahan was the assistant secretary for countering weapons of mass destruction at the Department of Homeland Security, from 2023 to 2025.
Add as preferred source on Google Tags AI Andy Jassy Anthropic Artificial intelligence Claude Fable 5 Executive Order Fable 5 MythosCopyright 2026 Nexstar Media Inc. All rights reserved. This material may not be published, broadcast, rewritten, or redistributed.
Comments: Link copiedMore Opinions - Technology News
See All
Opinions - Technology The hidden rebellion against returning to the office by Gleb Tsipursky, opinion contributor 1 day ago Opinions - Technology / 1 day ago 
