What happened in the Anthropic outage, by date?
- 9 June 2026: Anthropic released Claude Fable 5 and Claude Mythos 5.
- 12 June: The US Department of Commerce required access to be closed for all foreign nationals, including the company’s own staff. Anthropic could not separate them technically, so it paused the models for everyone, on AWS, Google Cloud, Microsoft, and its own API. This happened three days after launch.
- 26 June: About 100 US critical-infrastructure organizations received access to Mythos 5.
- 30 June: The restrictions were lifted.
- 1 July: Fable 5 opened again to users worldwide. The conditions were security checks and coordination with the state on future models.
In total, customers on every cloud received errors instead of answers for 19 days. According to MarketScale and Al Jazeera, banks, clinics, and software companies lost part of their working systems with no warning.
What did companies learn from the outage?
Supplier contracts did not expect that a service could be switched off by a government decision on all platforms at once. Companies whose agent was written for one model had to stop their processes. Companies where the model was a replaceable part could switch to another one.
The lesson is simple. A model is a dependency, and you need to be able to change it as you can change a hosting provider.
How do you build an agent with a replaceable model?
Keep the agent’s rules, its connections to CRM, phone system, and messengers, and its knowledge baseAI knowledge base. An AI knowledge base is a set of selected and prepared company documents that an AI assistant searches for answers: rules, instructions, a product catalog and answers to common questions. apart from the model. A knowledge base is a set of selected company documents that an AI assistantAI assistant. An AI assistant is a helper built on a neural network that, at a person’s request, answers questions, prepares texts and does simple tasks in work programs. searches for answers.
- We choose a model per process by answer quality on your tasks, by the place where it runs, and by price.
- We can replace the model without rewriting the rules and connections.
- An open-weights model is a neural network whose parameter files you can download and run on your own hardware. It is one option when you want full control over where the model runs.
- After rollout, the system, the data, and the access rights stay with you. If a model vendor or we disappear, the agent stays with you and can run on another model.
What should you check about your own agent?
Ask your vendor five things.
- Which model does the agent run on, and what happens if it becomes unavailable tomorrow?
- Where are the rules, the conversation history, and the knowledge base stored: with you, or inside someone else’s platform?
- Can you export all of it?
- Can you run it on another model, and in how many days?
- Who owns the code?
If there is no answer, you do not have an agent. You have a subscription to someone else’s service with your data inside.
Why is this not news in Russia?
Russian business has lived without direct access to the services of American developers for several years, so companies here have long worked through Russian models, open-weights modelsopen weights model. An open weights model is a neural network whose parameter files you can download and run on your own hardware. on their own servers, and intermediaries. What shocked Europe in June 2026 is a working norm for Russian companies. Law 243-FZRussia AI law. Russia’s AI law is Federal Law No. 243-FZ on supporting the development of artificial intelligence technologies, which introduces the concepts of large foundation models, sovereign models and national models. fixes the direction: sovereign and national models, and data in Russian data centres. A company whose agent is already built this way is ready for the next restriction from any side.
Personal data is processed on servers in Russia under 152-FZ. The lesson is the same for everyone: the model must be a replaceable part, and the agent’s rules and data stay with you.