US forces Anthropic to disable 2 AI models over jailbreak that exposed software flaws
Key Takeaways
- government’s unprecedented export control order on Anthropic’s Fable 5 and Mythos 5 models highlights a new frontier in cybersecurity: direct restriction of AI tools capable of vulnerability discovery.
- This raises critical questions for researchers about the balance between national security and the open research needed to harden global software.
Key Intelligence
Key Facts
- 1Anthropic received a U.S. government export control directive ordering it to suspend access to its Fable 5 and Mythos 5 AI models for all foreign nationals.
- 2The government cited a potential jailbreak that could allow Fable 5 to identify software vulnerabilities, providing only verbal evidence of a narrow, non-universal flaw.
- 3Anthropic will “abruptly disable” both models for all users globally, affecting hundreds of millions, rather than implementing a foreign-only block.
- 4The action follows a prior dispute in 2026 where Anthropic refused U.S. military use of its AI for surveillance and autonomous weapons, resulting in a supply chain blacklist.
- 5This marks the first time U.S. export controls have directly restricted access to AI models, shifting from a focus on hardware chips to software-level controls.
- 6Anthropic had called for greater government oversight of AI days earlier, including the ability to block model deployment, but now contests the scope of this order.
We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people.
In response to U.S. government order
Analysis
- Prevents foreign adversaries from exploiting AI vulnerability discovery
- Strengthens U.S. national security posture
- Disrupts legitimate cybersecurity research and vulnerability patching
- Creates uncertainty for AI model deployment globally
- Opaque government process may lead to overreach
Analysis
For cybersecurity professionals, this incident is a wake-up call: the U.S. government has moved from controlling AI hardware to directly throttling AI software that can be weaponized for vulnerability discovery. The order against Anthropic, based on an alleged jailbreak in its top-tier model, could signal a wave of restrictions that force the industry to rethink toolchains, vendor dependencies, and even the future of AI-assisted security research.
Anthropic’s abrupt announcement on June 13, 2026 that it will disable its most advanced AI models—Fable 5 and Mythos 5—for all users is a watershed event at the crossroads of artificial intelligence, national security, and cybersecurity. The company acted after the U.S. government issued an export control directive ordering it to suspend access to these models for all foreign nationals, citing national security concerns. According to Anthropic, the government believes there exists a “jailbreak” method capable of bypassing Fable 5’s safeguards, potentially enabling the model to identify software vulnerabilities. This is the first known instance of the U.S. leveraging export controls not on the hardware that powers AI—such as GPUs and advanced chips—but directly on access to AI software itself, marking a major policy shift with immediate global repercussions.
Anthropic’s abrupt announcement on June 13, 2026 that it will disable its most advanced AI models—Fable 5 and Mythos 5—for all users is a watershed event at the crossroads of artificial intelligence, national security, and cybersecurity.
The directive escalates a fraught relationship between Anthropic and the Trump administration. Earlier in 2026, Anthropic refused to allow the U.S. military to use its AI for domestic surveillance and fully autonomous weapons systems, leading to the company being placed on a supply chain blacklist set to take effect later this year. Now, the government has wielded its export authority to block foreign access, framing it as necessary to prevent adversaries from exploiting AI for vulnerability discovery. Anthropic, however, has pushed back strongly, stating it received only “verbal evidence of a potential narrow, non-universal jailbreak” and arguing that such a limited finding does not justify recalling a commercial model used by hundreds of millions of people. The company’s decision to disable the models globally—rather than implementing a complex geofencing solution—means that U.S.-based users are also suddenly cut off, amplifying disruption.
From a cybersecurity standpoint, the core of the dispute revolves around the dual-use nature of AI-assisted vulnerability detection. The capability to automatically identify software flaws is invaluable for defensive security teams seeking to patch systems before attackers exploit them. Yet, in the hands of nation-state adversaries or criminal groups, the same capability could accelerate zero-day discovery and weaponization. The U.S. government’s opaque process—verbal evidence with no written report—raises serious questions about due process and the risk of overreach in future AI regulations. Security researchers now face a murky landscape: tools that could have strengthened global software security are suddenly off-limits, potentially making the Internet less safe if the very means to find and fix bugs are restricted.
What to Watch
The global impact is immediate. With hundreds of millions of users suddenly denied access, developers and businesses that relied on these models for legitimate purposes—including vulnerability research, code review, and security automation—are left scrambling. The incident sets a precedent that other AI companies may face similar orders, especially those with advanced reasoning or coding capabilities. It also highlights a fundamental irony: just days earlier, on June 10, Anthropic had called for greater U.S. government oversight of AI, including the ability to block model deployment. Now it finds itself on the receiving end of that very authority, but in a manner it deems disproportionate. This underscore the unpredictability of regulatory action when there is no clear standard for assessing risks like jailbreaks.
The cybersecurity community will be watching closely to see if the government provides more transparency, and whether this signals the beginning of a new era of AI export controls that could reshape the global vulnerability research ecosystem. For organizations dependent on AI-driven security tools, the lesson is clear: reliance on a single provider’s models carries geopolitical risk, and a push toward open-source or locally deployed alternatives may accelerate. As AI capabilities continue to advance, the tension between innovation, national security, and openness will only intensify—and this incident may be just the first of many such confrontations.
Sources
Sources
Based on 3 source articles- lbc.co.ukAnthropic disables top - tier AI models after US government orders limiting foreign access over national security concernsJun 13, 2026
- asia.nikkei.comAnthropic disables top - tier AI models after US order limiting foreign accessJun 13, 2026
- arynews.tvAnthropic disables top - tier AI models after US order limiting foreign accessJun 13, 2026
From the Network
2 Top AI Models Disabled by Anthropic After US Jailbreak Order
The forced recall of Fable 5 and Mythos 5 highlights the tension between model safety and government overreach, based on an unverified jailbreak claim, and could reshape AI deployment norms.
StartupsAnthropic IPO at Risk as US Forces Disabling of 2 Flagship AI Models
The export control order against Anthropic creates regulatory uncertainty for AI startups and threatens its upcoming IPO, signaling that frontier AI models can be abruptly recalled on national securit
How we covered this story
Every story in our cybersecurity coverage is assembled from multiple primary sources, cross-referenced for factual consistency, and scored along three independent dimensions: sentiment, operational impact, and source-cluster confidence. Single-source rumors and unverifiable claims do not pass our editorial gate. When a story shows "Verified by N sources" with N≥2, the development is independently corroborated; when N=1, we mark it explicitly so readers can weigh the signal accordingly.
Impact scoring uses a 1-10 scale weighted toward regulatory, financial, and operational consequence rather than coverage volume. A topic that runs in every outlet but moves no real decisions ranks lower than a niche regulatory filing that reshapes how operators in the cybersecurity space have to behave. Read our full methodology for the scoring rubric, our glossary for term definitions, and our trends index for the longitudinal view across the beat.
| Signal on this page | What it tells you |
|---|---|
| Verified by N sources | Independent corroboration count. N≥2 is our confidence floor; N=1 is marked explicitly. |
| Impact score (1-10) | Regulatory + financial + operational weight. 8+ signals an experienced-operator action item. |
| Sentiment | Five-tier classification trained on labeled cybersecurity-specific corpora. |
| Timeline | Where applicable, the related-events sequence that contextualizes today's development. |