Claude Fable 5 had been available for just three days when Anthropic was forced to suspend access in mid-June 2026. The reason was a regulatory decision in the United States, not a technical problem. Anthropic itself had no influence over it. AI is not simply software that a company buys and then keeps forever, even if that would be convenient. Access to a model depends on decisions that customers cannot control. Those decisions may come from the provider, from a government or from a market that suddenly moves in a different direction.
Not every risk comes from outside
Fable 5 shows how access can disappear because of an external decision, even when neither the provider nor the customer can do anything about it. There is another risk that is just as real, but less visible. A provider can shape its offering in a way that gradually leaves customers with no practical way back.
Many large on-premises systems that companies have used for years are currently going through exactly this transition. The existing version is no longer being developed, new functionality is only available in the cloud and, sooner or later, support for the installed system also ends. In practice, customers are left with little choice but to move to the cloud.
ERP, data integration and AI from a single provider
Large cloud ecosystems now usually offer more than an ERP system. Data integration runs through the provider’s own platform, while AI is accessed through its own assistant. Each decision makes sense on its own. The components work well together. Integration is seamless. The company continues to deal with the same supplier.
The problem only becomes visible with some distance. A company that combines ERP, data integration and AI access with a single provider also ties all three to the same roadmap. Later, the company may want to handle data integration differently. It may want to switch to another AI model because it performs better or costs less. But the architecture stands in the way. The individual components were not designed to be replaced independently.
What starts as a convenient all-in-one solution becomes a dependency that is hard to undo. This affects any company that ties its digital infrastructure to a single ecosystem. It does not matter where the provider is based or whether the trigger is a regulatory decision, a change in the provider’s strategy or a substantial price increase.
A question for the board
This is not an issue that should be left to the IT department alone. It affects how much freedom a company will still have in five or ten years if a provider changes its terms, roadmap or technological priorities. Executive management and the board need to address it, not just the CIO.
Keeping control of data integration and AI
Companies should build their architecture so that they remain free to switch providers, add new components and reconsider earlier decisions. At Datalizard, we use our platform to support this approach. It connects data sources independently of the ERP system and uses its MCP interface to provide access to different AI models. Depending on its business, regulatory or strategic requirements, a company can work with Copilot, ChatGPT, Claude, Gemini or a self-hosted model. It can also make a different choice later.
The company remains in control of which AI can access which data and which tools it is allowed to use. Permissions, business rules and process logic also remain outside the AI model. They are managed in a controlled layer and do not need to be rebuilt when the model changes. Neither data integration nor AI access is then tied to the ecosystem in which the ERP system operates.
A real plan B does not simply mean choosing a second provider as a backup. It starts with an architecture in which data, business logic and the AI model are not inseparably linked.
Lizard Learning:
Before making the next cloud or AI decision, one question is worth asking: Can we replace this provider later without rebuilding our data integration, business logic and permission structure?
If the answer is no, the risk is already built into the architecture.
Learn more about MCP
datalizard Plattform
Understanding integration
