Anthropic Requires 30-Day Data Retention for Fable and Mythos Models

A support document quietly published by Anthropic reveals that its Fable and Mythos-class models come with a mandatory 30-day data retention period — a policy detail that enterprise customers negotiating data minimization agreements will want to read carefully.

Written by Lena Armitage · Bureau Tech · Jun 20, 2026

What the Policy Says

Anthropic's support documentation for its Mythos-class models — a product family that includes Fable — states plainly that data is retained for 30 days. The page, published at support.claude.com, does not describe this as a configurable default. It describes it as a retention practice, which in policy language typically signals a floor, not a dial.

That distinction matters. Many AI platform providers allow enterprise operators to negotiate data handling terms, including shorter retention windows or zero-retention configurations for sensitive workloads. If Anthropic's 30-day window is non-negotiable at the platform level, that narrows the options for customers who need to demonstrate data minimization to auditors or regulators.

Why "Mythos-Class" Is Worth Defining

Anthropic's product naming has expanded considerably. "Mythos-class" appears to be an internal or marketing designation grouping certain models — including Fable — under a shared capability or deployment tier. The support documentation does not fully define the boundaries of this class, and Bureau has not independently confirmed which other models, if any, fall under the same retention rules. Readers should treat "Mythos-class" as a term of art whose scope is not yet fully public.

The Enterprise Compliance Angle

For most consumer or developer use cases, a 30-day retention window is unremarkable. Cloud providers routinely retain logs for 30, 60, or 90 days for debugging and abuse-prevention purposes.

The calculus changes for enterprise buyers in regulated sectors. A hospital system evaluating an AI model for clinical documentation, or a law firm considering AI-assisted contract review, will typically require that data not persist beyond the immediate transaction — or at minimum, that retention periods be contractually defined and auditable. A mandatory 30-day window, if it cannot be waived, could disqualify Fable and Mythos from certain procurement processes outright.

It is worth noting that Anthropic does offer enterprise agreements, and it is possible that custom data handling terms — including shorter retention — are available under those contracts. The support documentation does not address this, and Bureau has not received clarification from Anthropic at time of publication.

What We Don't Know

The source material here is thin: a single support page, surfaced via Hacker News, with limited surrounding context. Bureau cannot confirm:

- Whether the 30-day window applies identically across API access, Claude.ai, and enterprise deployments - Whether Anthropic offers contractual carve-outs for regulated-industry customers - What, specifically, is retained (full conversation transcripts, metadata, embeddings, or some combination) - Whether this policy is new or has been in place since Fable and Mythos launched

Those are not small gaps. Until Anthropic clarifies the scope, enterprises should treat the 30-day retention requirement as a confirmed baseline and seek written confirmation of any exceptions before signing agreements.

Key takeaways

Anthropic's official support documentation states that Fable and Mythos-class models carry a mandatory 30-day data retention requirement.
The policy appears to apply at the platform level, meaning operators cannot opt out of the retention window through standard API agreements.
This is a meaningful constraint for enterprise customers in regulated industries — healthcare, finance, and legal — where data minimization is often a compliance requirement, not a preference.
The source material surfaced via Hacker News; the underlying claim comes from Anthropic's own support documentation, which Bureau has linked directly.
It remains unclear from the available documentation whether the 30-day window applies uniformly across deployment types (API, Claude.ai, enterprise contracts) or whether exceptions exist under custom agreements.

FAQ

What are Fable and Mythos in Anthropic's product lineup?

Based on available documentation, Fable appears to be a model within what Anthropic calls its 'Mythos-class' — a grouping of models under a shared tier or capability designation. Anthropic has not published a comprehensive public definition of which models fall under this classification.

Does the 30-day retention policy mean Anthropic can read my conversations?

Data retention and data access are related but distinct questions. Retaining data for 30 days means it is stored; it does not by itself specify who can access it or under what conditions. Anthropic's broader privacy policy governs access terms, and enterprise customers should review their specific agreements.

Can enterprise customers negotiate a shorter retention window?

The support documentation does not address this. It is common for AI platform providers to offer custom data handling terms under enterprise contracts, but Bureau has not confirmed whether Anthropic does so for Fable and Mythos. Customers should ask explicitly during procurement.

Is a 30-day retention period unusual for AI model providers?

Not inherently — many cloud and AI providers retain data for 30 to 90 days for operational reasons. What makes this noteworthy is the apparent lack of a configurable opt-out, which is a feature some competitors offer and which regulated-industry customers often require.

Citations

Data Retention Practices for Mythos-Class Models — Anthropic SupportAnthropic requires 30-day data retention for Fable and Mythos-class models
Hacker News — Discussion thread surfacing Anthropic retention policySecondary source that surfaced the Anthropic support documentation for editorial review
Anthropic Privacy PolicyGoverning document for how Anthropic handles user and operator data across its products; relevant context for interpreting retention practices