TL;DR
Anthropic disabled Fable 5 and Mythos 5 worldwide after a June 12 U.S. export-control directive barred access by foreign nationals, while OpenAI’s February GPT-4o retirement showed a provider can also end access through product policy. The confirmed events support a narrower point: users of hosted AI models depend on revocable access unless they control the model or have tested fallbacks.
Anthropic disabled its Fable 5 and Mythos 5 models worldwide after receiving a June 12 U.S. export-control directive barring access by foreign nationals, a shutdown that, paired with OpenAI’s February retirement of GPT-4o, has turned model access into an immediate operational risk for businesses and developers that build on hosted AI.
Confirmed: The directive required Anthropic to block use by any foreign national inside or outside the United States, including its own foreign-national employees, according to Business Insider. Anthropic said the letter arrived at 5:21 p.m. Eastern and did not give specific technical details; it said a global shutdown was the only workable way to comply.
Claimed or disputed: The administration’s stated basis was national security. Officials and allies of the White House later said the concern involved an alleged jailbreak in Fable 5 that could expose Mythos-class cyber capabilities. Anthropic disputed that account, saying the issue was narrow and did not give users novel capabilities beyond other public models, according to reports from Tom’s Hardware and The Verge.
Interpretation: OpenAI’s case was not a government action. The company retired GPT-4o and other older models from ChatGPT on February 13 after roughly two weeks’ notice, with reports saying it cited low active use and a shift toward newer systems. OpenAI’s API deprecation page says shutdown means a model or endpoint is no longer accessible. The point drawn by Thorsten Meyer AI’s Control Series is interpretive: both events show that customers access hosted models under rules set by governments and providers, not by the customers themselves.
The Switch: You Never Owned It
In 2026 a government turned off a frontier model worldwide in ~90 minutes — and a company retired a beloved one with ~2 weeks’ notice. You don’t own the model you build on. You access it. Access can be revoked.
Access is the only chokepoint that flips in an afternoon — and the version that hits you won’t be Washington, it’ll be a deprecation. Open weights you host can’t be deprecated, geofenced, repriced, or revoked. Short of that: route through a provider-agnostic gateway, keep a tested fallback, and treat every model string as a dependency that will be pulled.
API Dependence Becomes Business Risk
For companies building products on hosted AI, the risk is not limited to accuracy or latency. A model identifier embedded in production software can become a failure point when a provider retires a model, changes price, limits geography, lowers rate limits or updates behavior in a way that alters prompts and outputs.
The Anthropic order adds a state-power layer. Export controls were designed around goods such as chips and equipment, but this case shows they can also reach live AI services. That matters for security teams, banks, public agencies and software vendors whose processes may rely on one model’s capabilities, policy boundaries and response style.
AI model backup and fallback solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Two Routes to Lost Access
The Anthropic shutdown followed the release of Fable 5 and limited partner access to Mythos 5, which reports described as tied to advanced cybersecurity work. The government’s order did not shut down all Anthropic services; other Claude models were reported to remain available.
OpenAI’s GPT-4o retirement followed an earlier user backlash in 2025, when the company restored access after removing the model during a newer model rollout. In 2026, OpenAI moved ahead with the retirement after saying usage had fallen to about 0.1 percent of users, according to Business Insider and other reports.
“did not provide specific details of its national security concern”
— Anthropic, in a statement reported by Business Insider
AI model control and hosting hardware
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Security Evidence Remains Disputed
The government has not publicly released a full technical record supporting the Anthropic directive, and outside researchers have not reviewed the complete evidence behind the national-security finding. It is still unclear how long the controls will stay in place, whether a patch would satisfy U.S. officials, or whether similar orders could reach rival models.
For OpenAI, the open question is user impact rather than legality. Newer models may replace many GPT-4o tasks, but businesses and individual users still have to test whether prompts, support workflows, accessibility uses and companion-style interactions behave the same after migration.
The GPT-4 Millionaire: Future of Business Featuring Microsoft 365 Copilot: How to Leverage AI Language Models to Grow Your Company and How AI-driven Language Models Will Revolutionize the Way We Work
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Fallback Planning Moves Forward
Anthropic is expected to keep seeking a path to restore Fable 5 and Mythos 5 while addressing the alleged safety concern. Any agreement could shape how Washington applies export controls to hosted frontier models.
Customers that depend on AI APIs are likely to review provider-agnostic routing, fallback models, notice periods, pricing exposure and whether self-hosted open-weight models are needed for high-continuity workloads. The next test will be whether businesses treat model strings like external dependencies that can disappear.
Domain-Specific Small Language Models: Efficient AI for local deployment
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why did Anthropic disable Fable 5 and Mythos 5?
Anthropic said it received a U.S. export-control directive barring access by foreign nationals. Because that covered users inside and outside the United States, including Anthropic’s own foreign-national employees, the company said it disabled the models worldwide to comply.
Was GPT-4o retired for national-security reasons?
No. OpenAI’s GPT-4o retirement was a provider decision, not a government order. Reports said OpenAI cited low active use and the availability of newer models, while some users objected because they relied on GPT-4o’s behavior and tone.
Do businesses own the AI models they use through an API?
Usually no. They may own their application code, prompts, customer data and contracts, but the hosted model is controlled by the provider. Access can change through deprecation, pricing, regional limits, rate limits, policy updates or government orders.
Can open-weight models reduce this access risk?
They can reduce provider-deprecation and geofencing risk if a company hosts them itself. They also require compute, security work, monitoring and model operations that many teams currently outsource to API providers.
What should developers change now?
Avoid hardcoding one model string without a tested fallback. Track provider shutdown notices, run regression tests against alternate models, and design prompts and evaluations that can survive a model swap.
Source: Thorsten Meyer AI
