Microsoft has restricted employee access to Anthropic’s newly launched Claude Fable 5 as it reviews whether the model’s data-retention rules could create risks for customer data and confidential company information.
According to The Verge, the restriction comes even as Microsoft makes Claude Fable 5 available to GitHub Copilot and Microsoft Foundry customers. Rather than imposing a direct block, the company has kept the model out of the internal GitHub Copilot model picker used by its own employees.
Retention rules put internal rollout on hold
Microsoft’s legal teams are reviewing Anthropic’s updated retention requirements before deciding if Claude Fable 5 can be cleared for employee use.
The concern is that internal prompts involving customer data, proprietary code, product plans, or other sensitive material could fall under Anthropic’s new storage rules.
Anthropic’s data policy for Claude Fable 5 requires prompts and outputs to be retained so the company can operate the safety classifiers built into the model.
In most cases, that data is deleted after 30 days, but some prompts and outputs may be stored for up to two years if they are flagged as violating Anthropic’s usage policies.
That retention window has created uncertainty inside Microsoft, particularly because employees often use AI tools in technical workflows that may involve confidential code, customer-linked information, or unreleased product work.
Fable 5 debuts with tighter guardrails
Anthropic introduced Claude Fable 5 as its first Mythos-class model, positioning it as its strongest broadly available system for coding, research, vision, knowledge work, and long-running autonomous tasks.
The company also launched Claude Mythos 5 for a limited group of cyber defenders through Project Glasswing.
To reduce misuse, Fable 5 relies on safety classifiers that identify sensitive requests in areas such as cybersecurity, biology, chemistry, and model distillation. Those requests can be routed away from Fable 5 to a lower-risk Claude model, allowing Anthropic to offer broader access while limiting use cases it considers more dangerous.
Anthropic has said retained data from Fable 5 and similarly capable future models will not be used to train Claude or for non-safety purposes.
Still, Microsoft’s cautious approach shows how even limited data retention can complicate enterprise use when advanced AI systems are placed inside environments that handle confidential, regulated, or customer-related information.
