Microsoft suspends internal access to Anthropic’s Claude Fable 5 over data retention risks
The tech giant has removed the model from its internal workforce tools while external customers retain access, highlighting growing tensions between AI safety protocols and corporate data security standards.
Microsoft has restricted internal employees from using Anthropic’s Claude Fable 5 model, citing legal concerns regarding the AI provider’s new data retention policies. The decision excludes the model from the internal model picker used by staff, despite its availability to external customers via GitHub Copilot and Foundry. This move underscores the challenges large technology firms face when integrating advanced artificial intelligence tools that require data storage for safety monitoring.
The restriction stems from Anthropic’s implementation of new requirements to support its safety classifiers. Under the updated policy, the company retains user prompts and outputs for up to 30 days. Furthermore, data flagged as violating Anthropic’s usage policy may be stored for a period of up to two years. This stands in stark contrast to other Claude models currently available for internal use at Microsoft, which operate under Zero Data Retention rules.
Microsoft’s legal teams are currently evaluating these changes to determine if the model can be cleared for internal use. Internal communications have indicated that the primary concerns revolve around the handling of customer data and confidential information. It remains unclear whether the legal review will ultimately result in the model being reinstated for employee use.
Claude Fable 5 marks Anthropic’s first broad release from its Mythos class of AI models. The release follows recent statements from the company indicating that the family of models was initially deemed too capable in cybersecurity tasks to be released publicly. Prompt safeguards have since been implemented to mitigate these risks, though the resulting data retention requirements have created friction with Microsoft’s internal security standards.
Microsoft did not respond to requests for comment prior to publication. The situation highlights the complex regulatory and legal landscape surrounding enterprise AI adoption, where safety mechanisms often conflict with strict corporate data governance protocols.
