TL;DR
A June 12 U.S. export-control directive forced Anthropic to disable its Fable 5 and Mythos 5 models worldwide on roughly 90 minutes’ notice, according to the source material. Weeks earlier, OpenAI retired GPT-4o from ChatGPT and set API shutdowns, showing that both governments and providers can cut off AI systems that customers depend on.
A U.S. export-control directive on June 12 forced Anthropic to disable its Fable 5 and Mythos 5 models worldwide on roughly 90 minutes’ notice, according to source material from Thorsten Meyer AI, while OpenAI’s earlier retirement of GPT-4o from ChatGPT showed that providers can also remove widely used models through routine product decisions.
The Thorsten Meyer AI report, part of its Control Series, frames model access as a direct chokepoint for companies and users that rely on hosted artificial intelligence systems. The report says the Anthropic directive suspended access to Fable 5 and Mythos 5 by foreign nationals inside and outside the United States, including Anthropic’s own foreign-national employees. Because of that scope, the company disabled both models worldwide, according to the report.
The report attributes the directive to national-security concerns but says the detailed rationale was not provided in the letter described by Anthropic. It also says talks with the White House were expected after publication and that the controls remained in effect at the time of writing.
A separate access shock came from OpenAI’s retirement of GPT-4o and related models from ChatGPT, followed by scheduled API shutdowns. The report characterizes that change as a product and infrastructure decision, not a government action. For customers with production systems tied to the old model name, the practical result was a deadline to migrate before requests to the retired model began returning errors.
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.
Access Can End Instantly
The two episodes matter because they show that customers generally do not own the hosted models they build products, workflows, and internal systems around. They access them under terms set by a provider and, in some cases, rules imposed by a government.
That distinction can affect business continuity, security work, customer support, software features, and internal automation. If a model is withdrawn, blocked in a region, repriced, rate-limited, or changed in behavior, the effects can reach users and developers without the long lead times usually associated with infrastructure changes.
The report’s central claim is that access is the fastest AI chokepoint because it can be changed in a day, or faster. Compute, energy, data, and chip supply can constrain AI systems over time. Model access can fail immediately if an API endpoint is disabled or a model string stops resolving.
ORICO HS500 NAS Storage, AI Album and Virtual Machine, Flexible RAID Setup, 5* HDD Slot, 2 * M.2 NVMe SSD Slot, 2 * 2.5 GbE, 4K HDMl, Network Attached Storage (Diskless)
Local private cloud: Up to 136TB, 5* HDDs (24TB*5) + 2* NVMe SSDs (8TB*2), You can add more…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Two Switches, Same Risk
The Anthropic case is presented as the government version of the risk. Export controls were historically designed for physical goods and hardware supply chains, but the report argues that applying them to deployed AI models can function like a remote shutoff for software served by a U.S. company.
The OpenAI case is presented as the provider version. According to the report, OpenAI had previously tried to retire GPT-4o in 2025 and reversed course after user backlash. In 2026, with daily GPT-4o use reportedly reduced to a small share of activity, the retirement moved ahead.
The report also lists other ways access can change: geographic restrictions, pricing changes, rate limits, terms changes, and silent model updates that alter behavior. Those actions may be justified by legal, security, cost, or product reasons, but they can still break systems built around a specific model.
“You don’t own the model you build on. You access it.”
— Thorsten Meyer AI report
Local LLM Inference Optimization: A Comprehensive Guide to Quantization, Hardware Acceleration, and Efficient Private AI Deployment
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Security Rationale Still Disputed
It is not yet clear from the source material what specific national-security concern led to the June 12 directive affecting Anthropic’s Fable 5 and Mythos 5 models. The report says the letter gave no detailed rationale, and the merits of the concern remain contested.
It is also unclear how long the Anthropic controls will remain in place, whether exemptions will be granted, and how U.S. agencies may handle similar hosted-model controls in future cases. For OpenAI’s retirements, the remaining uncertainty is less about authority and more about operational planning: how much notice providers will give and how much behavior will change when customers migrate.
Agentic Coding With Claude Code: A Zero-Jargon Blueprint Helping Developers Automate Software Creation, Build Applications Faster, and Ship Projects (Jargon-Free Manuals)
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Builders Face Migration Deadlines
The next step for affected Anthropic customers is any outcome from talks with the White House and any clarification of the export-control scope. For OpenAI customers still tied to retired models, the next step is migration to supported models before shutdown dates turn into failed API calls.
The report recommends provider-agnostic routing, tested fallback models, and treating model names as dependencies that can expire. It also says open-weight models hosted by the user reduce the risk of deprecation, geofencing, repricing, or external revocation, though they carry their own operating costs and maintenance duties.
API ACCU-CLEAR Freshwater Aquarium Water Clarifier 8-Ounce Bottle
Contains one (1) API ACCU-CLEAR Freshwater Aquarium Water Clarifier 8-Ounce Bottle
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What happened to Anthropic’s Fable 5 and Mythos 5 models?
According to the source material, a June 12 U.S. export-control directive forced Anthropic to disable both models worldwide on roughly 90 minutes’ notice.
Was OpenAI’s GPT-4o retirement the same kind of event?
No. The report describes OpenAI’s GPT-4o retirement as a provider product decision, while the Anthropic shutdown was tied to a government directive. The risk for users was similar: access to a depended-on model ended or was scheduled to end.
Are the national-security reasons confirmed?
The directive was described as based on national security, but the source material says the detailed rationale was not provided in the letter and remains disputed.
What can developers do to reduce this risk?
The report recommends avoiding hard dependence on one hosted model, using provider-agnostic gateways, keeping tested fallback models, and planning migrations before model retirement dates arrive.
Source: Thorsten Meyer AI
