Microsoft told employees to steer clear of Anthropic’s Claude Fable 5, just one day after the model launched. The reason: Anthropic now stores user prompts and outputs for 30 days, and Microsoft’s lawyers aren’t comfortable with that.

The restriction, announced on June 10, 2026, applies specifically to Anthropic’s Mythos-class systems, which include the newly released Fable 5.

What changed at Anthropic

Fable 5 launched on June 9, 2026, positioned as a more accessible version of Anthropic’s Mythos 5 model. Alongside the launch, Anthropic updated its data retention policy to require a mandatory 30-day storage window for all prompts and outputs generated through its Mythos-class systems.

Anthropic frames this as a safety measure. The company says the retention window helps it combat emerging threats, including novel attacks and jailbreak attempts targeting its AI systems. The stored data is automatically deleted after 30 days unless it’s flagged for safety investigations or legal purposes. Anthropic has also stated the retained data will not be used for model training.

Previous enterprise-tier AI models typically offered zero data retention options, meaning conversations vanished the moment they ended. Anthropic’s new policy represents a notable departure from that norm.

Why Microsoft drew the line

Microsoft’s legal teams are currently reviewing the implications of Anthropic’s updated policy. The restriction appears precautionary rather than punitive.

Microsoft employees using Fable 5 for work-related tasks would effectively be sending proprietary Microsoft information to Anthropic’s servers, where it would sit for up to 30 days. Without clear definitions and contractual guardrails around what “safety investigations” and “legal purposes” mean in practice — the two exceptions Anthropic carved out for data that might otherwise be deleted after 30 days — Microsoft’s lawyers are right to want clarity before granting blanket access.

The bigger picture for enterprise AI

Anthropic’s decision to mandate retention essentially asks corporate users to accept a trade-off: better safety monitoring in exchange for less control over their data. Previous enterprise-tier AI models typically offered zero-retention guarantees, making this a significant shift for corporate buyers handling sensitive information.

Anthropic built its brand on safety and responsible AI development. The same safety-first philosophy that attracts users is now creating friction with enterprise partners over data governance expectations.