The United States lacks a coherent regulatory framework for advanced AI
In the unfolding contest between technological ambition and national security caution, the United States government has granted Anthropic conditional permission to restore access to its most powerful AI model, Mythos 5, to a select circle of vetted cybersecurity firms. The partial lifting of a two-week-old ban signals neither full confidence nor full suspicion, but something more unsettled — a nation improvising its relationship with a technology it both fears and cannot afford to cede. The moment reflects a deeper truth: that the rules governing humanity's most consequential tools are still being written, often in real time, under pressure.
- A sweeping federal ban on Anthropic's Mythos 5 and Fable 5 models, triggered by discovered vulnerabilities in their safety mechanisms, sent shockwaves through the AI industry just two weeks before this partial reversal.
- The Trump administration's anxiety that advanced AI could be weaponized by malicious actors at unprecedented scale has created a climate of restriction that now shapes how even America's own companies can deploy their own technology.
- Commerce Secretary Howard Lutnick's letter to Anthropic confirmed that new safeguards had satisfied federal concerns — but only enough to unlock access for a narrow, pre-approved group, leaving the broader public rollout still in negotiation.
- Rival OpenAI simultaneously launched GPT-5.6 under nearly identical government-validated constraints, revealing that this is not a story about one company but about an entire industry now subject to ad hoc federal gatekeeping.
- Anthropic's fraught relationship with the administration — including a Pentagon contract cancellation and a 'supply chain risk' blacklisting over disputes about military AI use — adds a charged political dimension to what might otherwise appear a routine regulatory process.
- With no transparent criteria, no published framework, and ongoing court battles, the United States is racing to lead the world in AI while still unable to answer how it intends to govern the technology it is racing to build.
The US government on Friday granted Anthropic permission to restore limited access to Mythos 5, its most advanced AI model, but only for a carefully screened group of American cybersecurity firms and infrastructure providers. The move partially reversed a ban imposed just two weeks earlier, when federal authorities had ordered Anthropic to suspend access to both Mythos 5 and its less advanced model, Fable 5, after identifying weaknesses in the safeguards designed to prevent misuse.
Commerce Secretary Howard Lutnick informed Anthropic by letter that the company had adequately addressed the security concerns that prompted the June 12 suspension. Anthropic confirmed the development and said it would move quickly to restore access for approved providers, while continuing weekend negotiations with officials to expand the rollout and eventually return Fable 5 to the general public.
The restrictions reflect a broader unease within the Trump administration about sophisticated AI falling into dangerous hands — particularly the fear that such systems could be exploited to identify and attack cybersecurity vulnerabilities at scale. On the same day, OpenAI launched its GPT-5.6 model under a similarly restricted arrangement, with access granted only after government validation. Both cases trace back to an executive order signed by President Trump earlier in June, establishing a voluntary federal review process for advanced AI models before public release.
The regulatory picture, however, remains deeply unclear. The White House has offered almost no public detail about how the review process works or what criteria govern approval. That opacity is especially fraught given Anthropic's troubled relationship with the administration: the company has opposed using its technology for mass surveillance or autonomous weapons, positions that led the Pentagon to cancel its contracts with Anthropic and prompted the administration to designate the company a 'supply chain risk.' Anthropic has contested that designation in court and has already secured at least one early legal win.
What the episode ultimately reveals is that the United States has no coherent framework for governing its most powerful AI systems — even as it competes urgently with China for technological dominance. Restrictions are imposed, then partially lifted, then renegotiated in real time. The question of how America will actually govern transformative AI remains, for now, largely unanswered.
On Friday, the US government gave Anthropic permission to restore access to its most powerful artificial intelligence model, Mythos 5, but only to a carefully vetted group of American cybersecurity firms and infrastructure providers. This marked a partial reversal of a sweeping ban imposed just two weeks earlier, when federal authorities had ordered the company to shut down access to both Mythos 5 and its less advanced sibling, Fable 5, after discovering weaknesses in the safeguards meant to prevent the technology from being weaponized or exploited.
Commerce Secretary Howard Lutnick delivered the news in a letter to Anthropic on Friday, stating that the company had successfully addressed the security risks that had triggered the initial suspension on June 12. "I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model," Lutnick wrote. Anthropic confirmed the development in a statement, describing Mythos 5 as its strongest cybersecurity model and pledging to quickly restore access to the approved set of providers. The company also signaled its intention to continue negotiating with government officials over the weekend to expand the rollout further and eventually make Fable 5 available to the general public again.
The restrictions on Anthropic emerged from a broader anxiety within the Trump administration about advanced artificial intelligence systems falling into the wrong hands. There is genuine concern that sophisticated AI models could be weaponized by hackers and malicious actors to identify and exploit cybersecurity vulnerabilities at scale and speed. On the same day Anthropic received its partial reprieve, rival company OpenAI launched its GPT-5.6 model under a similarly restrictive arrangement, with access granted only on a case-by-case basis following government validation. Both moves reflect the implementation of an executive order signed by President Trump earlier in June that established a voluntary federal review process for advanced AI models, requiring companies to submit their systems for national security assessment before public release.
Yet the regulatory landscape remains murky. The White House has disclosed almost no details about how the review system actually works, which models fall under its jurisdiction, or what criteria determine approval. This opacity has become particularly consequential given the fraught relationship between Anthropic and the Trump administration. The company has publicly opposed allowing its technology to be used for mass surveillance or autonomous weapons systems—positions that led the Pentagon to cancel existing contracts with Anthropic. In response, the administration designated Anthropic a "supply chain risk," effectively blacklisting the company over disagreements about military applications of its AI technology. Anthropic has challenged that designation in court and has already won at least one early legal victory in the ongoing case.
The back-and-forth between Anthropic and federal authorities underscores a fundamental problem: the United States lacks a coherent, consistent regulatory framework for advanced artificial intelligence, even as it races to maintain technological superiority over China and other competitors. The current approach appears to be ad hoc—restrictions imposed, then partially lifted, then negotiated in real time—rather than grounded in clear rules that companies and regulators can both understand and follow. As discussions between Anthropic and the government continue, the question of how America will actually govern its most powerful AI systems remains largely unanswered.
Citas Notables
I have determined that appropriate safeguards are in place to permit certain trusted partners to access the Claude Mythos 5 Model— Commerce Secretary Howard Lutnick, in letter to Anthropic
We are pleased to see this progress and continue to work with the government to expand access to Mythos 5 and make Fable 5 available for general use again— Anthropic statement
La Conversación del Hearth Otra perspectiva de la historia
Why did the government shut down Mythos 5 in the first place if they were just going to let it back out two weeks later?
They found vulnerabilities in the safety mechanisms—ways the model could potentially be misused. But they also realized that completely cutting off access wasn't practical, especially for the cybersecurity firms that actually need tools like this to defend critical infrastructure. So they bought time to patch the problems.
And now it's back, but only for trusted partners. How does the government decide who's trusted?
That's the real question nobody's answering. The administration hasn't published the criteria. It seems to be happening behind closed doors, case by case, which is why Anthropic is still negotiating.
What's the tension with the Pentagon about?
Anthropic refused to let the military use their AI for autonomous weapons or mass surveillance. The Pentagon saw that as obstruction, so they canceled contracts and blacklisted the company. Now Anthropic is fighting that designation in court.
So the company is simultaneously trying to restore public access while being treated as a national security threat?
Exactly. They're in this strange position where they're cooperating with the government to prove they're trustworthy, but the government has also formally designated them as a supply chain risk. It's contradictory.
Does this happen to other AI companies?
OpenAI got similar restrictions the same day, but they seem to have a better relationship with the administration. The real issue is that there's no actual rulebook. Each company is negotiating its own terms with regulators who are making it up as they go.