TL;DR
A June U.S. export-control order forced Anthropic to disable Fable 5 and Mythos 5 worldwide, reports said, while OpenAI’s February retirement of GPT-4o from ChatGPT removed a model many users had built around. The paired events show that hosted AI access can be withdrawn by governments or providers, even when users’ own code and data remain in place.
A U.S. export-control order forced Anthropic to disable its Fable 5 and Mythos 5 AI models worldwide in mid-June 2026, while OpenAI’s February retirement of GPT-4o from ChatGPT showed a separate route by which widely used AI tools can disappear: a provider’s product decision. The paired events matter because businesses, developers and users often build workflows around hosted models they can access but do not own.
Confirmed by reports: Tom’s Hardware reported that the U.S. order barred access to Anthropic’s newest models by foreign nationals, including Anthropic employees, leaving the company to disable Fable 5 and Mythos 5 worldwide. Source: https://www.tomshardware.com/tech-industry/artificial-intelligence/us-export-control-order-forces-anthropic-to-disable-claude-fable-5-and-mythos-5-worldwide.
The order cited national security, according to the reports. Business Insider reported that White House officials were concerned that the models’ safeguards could be bypassed; Anthropic, according to that report, argued that no universal jailbreak existed. Source: https://www.businessinsider.com/why-white-house-ordered-export-controls-anthropic-mythos-fable-2026-6.
Separately, OpenAI retired GPT-4o from ChatGPT on February 13, 2026, after about two weeks’ notice, according to reports. OpenAI’s official deprecations page lists related API shutdowns, including chatgpt-4o-latest on February 17 and gpt-4o-2024-05-13 on October 23, 2026. Source: https://developers.openai.com/api/docs/deprecations.
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.
Hosted Models Became Operational Risk
For readers who use AI at work, the lesson is direct: hosted model access is a service relationship, not ownership. If an endpoint is cut off, a legal review, support workflow, coding assistant, research process or customer-facing product can fail even if the user’s data, prompts and application code remain intact.
For builders, this makes a model identifier a dependency that needs version control, fallback planning and testing. Provider decisions can retire models, change pricing, add regional limits or alter behavior. Self-hosted open-weight models reduce some of that exposure, but they add costs, security duties and performance tradeoffs.
Start Here! Learn Microsoft Kinect API
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Export Controls Meet Cloud AI
Export controls were designed for goods and technology that cross borders. Applied to hosted AI models served through cloud APIs, the same tool can act as a remote access cutoff. The Anthropic case is being watched because the order appeared to reach customers and employees outside the United States through a U.S.-based provider.
The OpenAI case was less dramatic but more familiar to developers: a product retirement. Business Insider reported that OpenAI said only 0.1% of users still operated GPT-4o before the February retirement, after an earlier removal attempt had drawn user backlash in 2025. Thorsten Meyer AI framed the two events as different hands on the same switch.
“You don’t own the model you build on. You access it.”
— Thorsten Meyer AI Dispatch
Building AI-Powered Products: The Essential Guide to AI and GenAI Product Management
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Rationale And Remedies Remain Thin
It is not yet clear how long the Anthropic controls will stay in place, whether Fable 5 and Mythos 5 will return, or what technical evidence led officials to seek a worldwide cutoff. It is also unclear what recourse affected customers have beyond waiting for Anthropic’s talks with the U.S. government.
For OpenAI users, the main open issue is not whether GPT-4o left ChatGPT; it did. The open question is how often similar retirements, regional limits, pricing changes or behavior changes will affect production systems built on other hosted models.
eufy Security HomeBase Professional S1, 4G LTE and 24h Battery Backup, Built-in 32 GB Storage and Expandable up to 16 TB, Advanced Local AI, Compatible with eufy Products, No Monthly Fee
Uninterrupted Video Security: Stay secure during power or Internet outages with 4G connectivity and a 24-hour built-in battery…
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Anthropic Talks And API Deadlines
Anthropic is expected to keep pressing U.S. officials for relief or a narrower compliance path. Customers relying on the disabled models will be watching for reinstatement, replacement access or guidance on alternate models.
Developers using deprecated OpenAI snapshots face fixed API dates listed in OpenAI’s deprecation schedule. The practical next step for AI-dependent teams is to inventory model dependencies, test fallback providers, keep prompts portable and treat every hosted model string as something that may be pulled on a deadline.
Generative AI on Kubernetes: Operationalizing Large Language Models
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
What actually changed for Anthropic users?
Reports said Anthropic disabled Fable 5 and Mythos 5 worldwide after a U.S. export-control order restricted access by foreign nationals. Users who depended on those models had to shift to other models or pause affected workflows.
Did OpenAI turn off GPT-4o for the same reason?
No. The Anthropic shutdown was tied to a government order citing national security. OpenAI’s GPT-4o retirement was a provider decision tied to product direction, usage and model support, according to reports and OpenAI’s deprecation schedule.
Does this mean users never own an AI model?
Users own their data, code and prompts only to the extent their contracts and systems allow. A hosted proprietary model is different: access runs through the provider’s infrastructure and can be limited, retired or changed.
How can companies reduce the access risk?
Companies can keep model calls behind an abstraction layer, test fallback models, avoid hardcoding one provider where possible and run drills before shutdown dates. Some may add self-hosted open-weight models for high-risk workflows, while accepting the added operational burden.
Are the Anthropic models coming back?
That remains unclear. Reports said Anthropic planned talks with U.S. officials, but no final timetable or reinstatement plan had been confirmed in the source material.
Source: Thorsten Meyer AI
