Key Points
- Anthropic disabled its newest AI models (Fable, Mythos) globally after the U.S. govt restricted access for foreign nationals.
- The order followed concerns the models could be “jailbroken” to expose software vulnerabilities, raised in part by Amazon CEO Andy Jassy.
- IANS Faculty say this signals export-style control over frontier AI, creating immediate availability, concentration and resilience risk for enterprises.
Anthropic Shuts Off Access to Flagship AI Models After U.S. Order
Anthropic abruptly shut down access to its most advanced AI systems -- Fable 5 and Mythos 5 -- after receiving a U.S. government order requiring it to block access to the models for all foreign nationals, including non-U.S. citizens inside the United States.
The directive, issued under export-control authorities, prohibited use of the models by any foreign individual regardless of location. Anthropic said it could not reliably enforce nationality-based access controls across its global platform, so it took the models offline entirely rather than attempt partial restrictions.
The government’s action followed alleged concerns that the models could be “jailbroken” to identify software vulnerabilities. Amazon researchers reportedly claimed that Fable 5 could be prompted to surface cyberattack-relevant information, and CEO Andy Jassy reportedly raised those findings directly with senior U.S. officials.
Officials pressed Anthropic to address the issue or withdraw the models. When the company reportedly did not move quickly to do so, regulators imposed export controls that restricted operation of the models under the new rules, leading to the full shutdown.
Anthropic pushed back, arguing the alleged “jailbreaks” revealed only narrow or already-known software vulnerabilities and warning the standard could slow deployment across the industry.
Big Picture
While it's unclear how long this restriction will last, frontier AI models are starting to operate like infrastructure governed by national policy. That shift means access can change suddenly, with immediate impact across companies that rely on the same models.
For companies that embedded these models, a production capability used for development and security work was removed overnight.
“The lesson is that you no longer fully control whether a capability you depend on will exist tomorrow. A vendor can pull it, and now we have proof that a government can force that pull, too.” George Gerchow, IANS Faculty.
Those models often sit underneath multiple tools across an environment. When Anthropic shuts them down, the impact spreads across workflows immediately.
“When five or six of your SaaS vendors all sit on the same underlying model, a single upstream event, such as a deprecation or a forced suspension at the provider, rolls across your whole portfolio at once.” George Gerchow, IANS Faculty.
Most organizations haven’t mapped where these dependencies exist. When access is cut off, multiple systems can lose a core capability at once with little warning.
“IANS clients need to ensure no business process relies on a restricted AI model that could, in turn, provide products or services to international companies or people that are banned by Department of Commerce guidelines. The problem is, how can organizations effectively identify their AI supply chains when everything is moving so quickly?” Aaron Turner, IANS Faculty.
IANS Faculty Recommendations
- Run this through model risk management (MRM): Apply existing MRM discipline to AI usage so this risk is governed alongside other model-driven decisions regulators already scrutinize.
- Map shared model dependencies: Identify where vendors rely on the same underlying model to expose where a single intervention creates multi-system impact.
- Contract for model transparency: Require disclosure of model changes, suspensions, or access restrictions; if vendors refuse, treat that as a control gap and mitigate.
- Design for portability at the model layer: Ensure applications can switch models without major rework so a forced shutdown doesn’t take workflows offline.
- Plan for government-driven removal scenarios: Add AI model loss to resilience planning with fallback options, escalation paths, and communication plans before it happens.
Authors & Contributors
Dan Maloof, Editor-in-Chief, IANS News
George Gerchow, IANS Faculty
Aaron Turner, IANS Faculty
Jeff Brown, IANS Faculty
Although reasonable efforts will be made to ensure the completeness and accuracy of the information contained in our News & blog posts, no liability can be accepted by IANS or our Faculty members for the results of any actions taken by individuals or firms in connection with such information, opinions, or advice.