US clears Anthropic’s Claude Mythos 5 for critical infrastructure following national security ban

The United States government has granted Anthropic regulatory approval to redeploy its highly advanced artificial intelligence model, Claude Mythos 5, for a select group of organizations responsible for operating and defending critical infrastructure[1][2]. This decision represents a major de-escalation in a tense standoff between the federal government and the prominent AI developer, which began when regulators abruptly blocked access to the company’s most powerful frontier models over national security concerns[3][4]. While the partial reinstatement marks a significant operational milestone for Anthropic, general access to its next-generation models remains restricted as negotiations continue[1]. The company is actively working with officials to establish a pathway for broader commercial access and to lift the ongoing suspension of its consumer-facing counterpart, Claude Fable 5, though no definitive timeline has been established[4][5].

The regulatory friction began shortly after the highly anticipated debut of Claude Mythos 5 and Claude Fable 5, two closely related systems representing the pinnacle of Anthropic's technical capabilities[6][7]. Positioned as next-generation frontier models, they demonstrated unprecedented performance in complex software engineering, advanced reasoning, and scientific research[6][7]. While Fable 5 was designed with conservative safety classifiers to restrict answers in high-risk domains for public use, Mythos 5 was engineered without these specific limitations to serve as a specialized, ultra-powerful tool for cybersecurity professionals and infrastructure defenders[6][7][8]. However, just weeks after their initial rollout, the U.S. Commerce Department intervened with an emergency export control directive that forced Anthropic to disable both models for all users worldwide[9][3]. The sudden shutdown followed warnings from major cloud providers and security analysts that the models could be jailbroken by malicious actors to construct sophisticated cyber-weapons or facilitate biological and chemical threats, prompting federal authorities to intervene under national security protocols[3][6][4].

The path to restoring access began through intensive, daily negotiations between Anthropic executives and federal officials, culminating in a formal agreement that outlines strict oversight and operational safeguards[4]. In a letter addressed to Anthropic Chief Compute Officer Tom Brown, U.S. Commerce Secretary Howard Lutnick cited significant progress in resolving the government's security concerns, noting that appropriate protocols had been established to permit trusted partners to utilize the model[4]. Under the terms of the newly approved arrangement, Claude Mythos 5 can be distributed to more than one hundred designated U.S. institutions, including prominent private corporations and vital government agencies[4]. Crucially, the federal government has waived traditional export licensing requirements for these pre-approved entities and their foreign national employees, as well as for Anthropic’s own international staff, removing an administrative bottleneck that had threatened to stall crucial cyber-defense operations[4].

Despite this breakthrough for critical infrastructure operators, the fate of Claude Fable 5 remains undecided, creating a complex landscape for Anthropic’s commercial ambitions[4][5]. Because Fable 5 was intended as a widely available commercial model, its continued suspension leaves standard enterprise and consumer subscribers relying on older systems like Claude Opus 4.8[6][7]. Although federal regulators did not formally address Fable 5 in their latest directive, sources close to the discussions indicate that both parties are moving toward a framework that could eventually restore the consumer model under enhanced guardrails[4][10]. The pressure to resolve these regulatory hurdles is compounded by intense industry competition[4]. On the exact same day that Anthropic secured its partial clearance for Mythos 5, its primary rival, OpenAI, launched its newest model, GPT-5.6, to a restricted group of government-approved partners, underscoring the relentless pace of development among top-tier AI labs[4].

The regulatory struggle also highlights deeper geopolitical tensions and the shifting relationship between Silicon Valley and Washington[10]. Anthropic has historically maintained a cautious stance toward militarized AI, previously drawing scrutiny from federal authorities for refusing to allow its models to be used by the U.S. military for domestic surveillance and autonomous weapons operations[10]. Some industry experts have expressed concern that overregulation and prolonged federal shutdowns of powerful domestic AI models could inadvertently undermine national security by delaying critical defensive updates[11]. These commentators argue that while U.S. developers face operational freezes, international competitors, particularly open-source developers in China, are free to advance their systems without similar constraints, potentially leaving Western digital infrastructure more vulnerable in the long run[10][11].

Ultimately, the conditional return of Claude Mythos 5 illustrates the complex and rapidly evolving frontier of artificial intelligence governance, where national security concerns must be carefully balanced against technological progress[3]. The agreement establishes a potential blueprint for how future highly capable models will be managed, favoring highly controlled, vetted access programs over outright bans[4][7]. As Anthropic works to expand the list of approved organizations for Mythos 5 and aims to reinstate Fable 5 for the general public, the tech industry will be watching closely to see if this collaborative approach can successfully mitigate extreme risks without stifling the innovation that drives the modern digital economy[5].