Phoenix 24
More than 100 approved organizations can again use the advanced cybersecurity model, while broader access remains suspended.
Washington, June 2026
Anthropic has begun restoring access to Claude Mythos 5 for a limited group of American organizations after the United States government partially relaxed restrictions imposed on the company’s most advanced artificial intelligence systems. The decision allows selected institutions responsible for protecting critical infrastructure to resume using the model, but it does not amount to a general release.
Mythos 5 is Anthropic’s most powerful model specifically designed for cybersecurity work. It can analyze complex software, identify vulnerabilities and assist specialists defending digital systems used by energy providers, telecommunications networks, financial institutions and other strategically important sectors.
The United States government authorized more than 100 trusted organizations to regain access, according to reports surrounding the agreement. The approved group reportedly includes major companies and institutions involved in critical infrastructure protection, several of which participate in Anthropic’s Project Glasswing cybersecurity initiative.
Anthropic said it was moving rapidly to reactivate the service for eligible organizations. The company also confirmed that negotiations with federal authorities were continuing as it seeks wider availability for Mythos 5 and the restoration of its more broadly oriented Fable 5 model.
The partial reversal follows an extraordinary government intervention that began on June 12. Anthropic disabled both Mythos 5 and Fable 5 after receiving an export-control order that restricted access to the models by foreign nationals, including some employees and contractors working inside the United States.
Because Anthropic could not easily separate every user according to citizenship and organizational approval, the company temporarily withdrew the systems more broadly. The shutdown disrupted customers, developers and cybersecurity teams that had begun integrating the models into sensitive technical workflows.
Federal authorities justified the restrictions through national security concerns. Advanced cybersecurity models can help defenders locate weaknesses before attackers exploit them, but the same analytical capabilities can potentially be misused to discover vulnerabilities, bypass safeguards or automate offensive operations.
Mythos 5 attracted particular scrutiny because of its ability to examine complicated software environments and identify security flaws at significant speed. The model had already been tested by technology companies and government institutions seeking to strengthen the protection of widely used operating systems, browsers and critical applications.
The renewed authorization does not eliminate those concerns. It creates a controlled exception for organizations considered sufficiently trusted and strategically important. Institutions outside the approved group remain unable to use the model, even when their work may involve legitimate cybersecurity research.
One important change concerns foreign nationals employed by approved American organizations. Under the initial order, access restrictions applied to them regardless of where they lived or worked. The new arrangement reportedly allows qualified non-citizen personnel inside authorized institutions to use Mythos 5 without requiring separate export licenses.
That exemption could reduce some of the operational disruption created by the original policy. American technology and cybersecurity companies rely heavily on international talent, and a citizenship-based restriction can divide teams whose members collaborate on the same software, infrastructure and security problems.
Fable 5 remains unavailable to the broader public. Although it shares important technical foundations with Mythos 5, Fable is designed for more general commercial and consumer applications. Reports indicate that the government may be approaching a decision on restoring it, but no definitive timetable has been announced.
The unequal treatment of the two models reflects their intended uses. Mythos 5 is being returned first because authorities consider its defensive cybersecurity value particularly important to selected American organizations. Fable 5 would reach a much larger and less controlled user population, making its reintroduction politically and technically more complicated.
The restrictions have produced criticism from civil liberties specialists, technology companies and AI policy researchers. One central concern is the absence of transparent criteria explaining how organizations are selected for access and why others are excluded.
When the government determines which private companies may use a powerful commercial technology, the process can create competitive advantages. Approved organizations gain access to capabilities that remain unavailable to rivals, researchers and smaller cybersecurity firms that may also contribute to national defense.
Critics have warned that this concentration of authority could allow government agencies to shape the AI market through confidential decisions rather than clearly published standards. Questions also remain about whether rejected organizations can appeal, how access may be withdrawn and what evidence is required to demonstrate trustworthiness.
Supporters of the restrictions argue that frontier AI systems demand exceptional precautions. A model capable of identifying software weaknesses at scale may produce benefits for defenders while simultaneously reducing the expertise required for malicious actors to conduct sophisticated attacks.
The policy debate therefore involves more than the conventional tension between innovation and regulation. It concerns whether powerful AI capabilities should be treated like ordinary software, controlled dual-use technology or strategic infrastructure requiring continuous government supervision.
Anthropic has cooperated with federal officials on additional safeguards and risk-management measures. Government representatives reportedly recognized progress made by the company in addressing the concerns that triggered the original suspension, although they reserved the ability to change the licensing conditions as the situation develops.
The episode has also affected competing AI developers. OpenAI reportedly faced similar government pressure surrounding access to advanced GPT-5.6 systems, with distribution limited to a narrow group of vetted partners. The parallel restrictions indicate that Washington is moving toward model-specific controls rather than focusing on a single company.
Technology firms have argued that such interventions could weaken American competitiveness if access becomes unpredictable. Cybersecurity teams may turn toward foreign or open-source alternatives when domestic models are suddenly suspended, potentially reducing the government’s ability to supervise how the technology is used.
The government faces the opposite risk if it permits unrestricted deployment. Once a model becomes widely available, controlling copies, derivatives and unauthorized access becomes increasingly difficult. Authorities are therefore attempting to preserve valuable defensive uses without allowing the most sensitive capabilities to circulate freely.
For Anthropic, the partial restoration represents progress but not a resolution. The company can again serve important cybersecurity partners, yet its two leading models remain subject to federal decisions capable of interrupting commercial deployment with little notice.
The Mythos 5 controversy may become an early example of how governments regulate frontier AI through access lists, licensing conditions and national security exemptions. Such arrangements can respond quickly to emerging risks, but they can also create uncertainty when their rules remain confidential.
More than 100 organizations now regain a powerful defensive tool. Thousands of other potential users remain outside the authorized circle, waiting for clearer standards or another government decision.
La inteligencia más poderosa también obliga a decidir quién puede utilizarla. / The most powerful intelligence also forces society to decide who may use it.