Microsoft Restricts Employee Access to Anthropic’s Fable 5 AI Over Data Retention Concerns

Microsoft has reportedly decided to limit employee access to Anthropic’s Claude Fable 5, the first publicly available Mythos-class artificial intelligence model, citing concerns related to data retention and compliance requirements.

The move may appear surprising given Microsoft's aggressive push toward integrating AI tools across its products and workplace ecosystem. With thousands of employees worldwide and significant investments in AI-driven productivity solutions, many expected Microsoft to fully embrace Anthropic’s latest flagship model internally.

However, according to a report by famous publication, Claude Fable 5 has not been made available to Microsoft employees through the company’s internal version of GitHub Copilot. This is despite Microsoft offering access to the Mythos-class model to customers through GitHub Copilot and Microsoft Foundry services.

The development follows reports that Microsoft recently discontinued Claude Code subscriptions for employees, with industry observers suggesting that operating costs may have contributed to that decision. Yet, in the case of Fable 5, financial considerations do not appear to be the primary factor.

The report states that Microsoft’s concerns are linked to Anthropic’s updated data retention practices for its Mythos-class models.

Under the new policy, prompts and outputs generated through Mythos-class systems are stored for 30 days. Anthropic says this retention period is necessary to support safety monitoring tools designed to identify new attack methods, jailbreak attempts, and other potential misuse of the technology.

Anthropic maintains that the stored information is deleted after 30 days and is not used to train future AI models. However, the company also notes that prompts or outputs flagged for safety investigations, legal obligations, or potential violations of usage policies may be retained for up to two years.

These changes have triggered compliance and legal concerns among enterprise customers, particularly organizations that handle confidential business information or sensitive customer data. Traditionally, many enterprise AI offerings have included zero-retention options, ensuring conversations are deleted immediately after use.

According to the report, Microsoft’s legal and compliance teams are currently reviewing Anthropic’s policy framework. The primary concerns reportedly revolve around safeguarding customer information and protecting proprietary corporate data. As a result, it remains unclear whether Claude Fable 5 will eventually receive approval for broader internal deployment within Microsoft.

Notably, Microsoft employees continue to have access to other Claude models internally because those offerings operate under zero data retention arrangements.

The decision has also sparked reactions across social media platforms, where users responded with a mix of humour and skepticism.

One person shared a video of a person going back home and wrote, “Microsoft devs on their way home after realising they can’t vibe code with Claude Fable 5 at work.”

Another user posted a scene from the movie Prisoners featuring actor Jake Gyllenhaal damaging items in an office, adding, “Microsoft employees have realised that they'll now have to really work (some of them don't know how).”

Not everyone accepted the official explanation. One user questioned Microsoft’s reasoning and suggested cost considerations may still be involved, commenting, “Bullshit excuse. They just don’t want to pay $50 per million tokens.”

As enterprise AI adoption accelerates, Microsoft’s cautious approach highlights the growing tension between access to cutting-edge AI systems and the compliance requirements that large organisations must navigate.