Anthropic and AWS Publish Different Rules for Fable 5
Anthropic's documentation says Claude Fable 5 and Claude Mythos 5 are Covered Models: zero data retention is 'not available for either model.' AWS's Bedrock documentation says its listed models include Fable 5, for which eligible customers may request full ZDR through their account team; the list does not include Mythos 5. This piece starts from which channel each document describes, then covers the 30-day rule that began on June 9, 2026.
本文另有中文版：Fable 5 能零資料保留嗎？Anthropic 與 AWS 的文件各寫了不同的答案
Can Claude Fable 5 run with zero data retention?
Anthropic’s documentation says Fable 5 and Mythos 5 are Covered Models, with zero retention “not available for either model.” AWS’s Bedrock abuse-detection page, after listing Fable 5 among other models, says eligible customers may request full ZDR through their account team. That list does not include Mythos 5.
Each sentence has its own subject and its own channel: the first is Anthropic’s rule for Covered Models, the second is how AWS writes it up for Bedrock. This piece does not rush to say who is wrong. It starts from which road each document is describing.
Set side by side, the documents describe different access conditions. Under Anthropic’s document, a ZDR-configured organization must enable retention for Covered Models; AWS separately describes an eligible-customer full-ZDR request for Fable 5.
The rule written into an error message
Fable 5 and Mythos 5 launched the same day, but Fable 5 is the generally available one. Mythos 5 is invitation-only, limited to approved customers inside something Anthropic calls Project Glasswing, built for defensive cybersecurity work, with no self-serve sign-up. Fable 5 is the one that went generally available on the Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry, all on June 9.
Anthropic’s announcement states a 30-day requirement for traffic to Mythos-class Covered Models across first- and third-party surfaces, as a change for organizations that had configured ZDR. AWS separately says eligible customers may request full ZDR for Fable 5. Anthropic says the retained data is used for safety work rather than model training, and that it logs human access.
The announcement leaves itself room, though. Deletion after 30 days happens “in almost all cases,” not in every case — an exception the company built into its own language and did not explain further.
Who this actually touches is buried in one qualifying sentence. It applies only to organizations that had already configured zero data retention: a Console workspace with ZDR set, Claude Code used inside Claude Enterprise with ZDR configured, or Claude reached through AWS Bedrock, Google Cloud’s Agent Platform, or Microsoft Foundry with ZDR enabled. Consumer plans are untouched. For every organization outside that list, nothing changes, because their data was already being retained under existing terms.
An organization whose retention configuration does not meet the requirement gets a 400 error instead of an answer: type invalid_request_error, message “In order to access this model, your organization or workspace must have data retention enabled.” That is Anthropic’s own API error text. It marks a rejected request; it does not show that the content of that request was retained.
The access condition depends on the channel and agreement.
Where the retained data actually sits
The processing and location wording differs by channel. Through the Claude API, including Claude Platform on AWS, Anthropic’s documentation says Anthropic itself handles the retained data. Through Amazon Bedrock and Google Cloud’s Agent Platform, the same documentation says retained data “stays within your cloud provider’s environment,” the cloud vendor’s environment.
AWS’s own Bedrock documentation describes that same arrangement differently. Fable 5 and Mythos 5 are models that “require provider data sharing” on Bedrock. A customer has to explicitly set their retention mode to provider_data_share before either model will even respond; leave it at none or default, and the model shows up as unavailable.
Under that mode, AWS writes that user prompts and completions are shared with Anthropic and retained for up to 30 days for trust and safety purposes.
Anthropic describes retained data staying in the cloud provider’s environment; AWS also describes sharing with Anthropic. Storage location and access are different questions. The documents available here do not fully reconcile the arrangements, and this article does not infer a physical transfer from sharing alone.
AWS’s Bedrock documentation also says that none is the zero-retention mode for ordinary inference; it must be set explicitly and is subject to separately stated CSAM handling.
There’s a second mismatch, the one this piece opened with: whether zero retention is available at all. Against Anthropic’s “not available for either model,” a separate AWS document, covering abuse detection on Bedrock, says: “For these models, eligible customers may request full ZDR through their AWS account team.” The models listed on that AWS page include Fable 5, not Mythos 5. Not any customer, and not automatically. Eligible customers, through a request.
The same AWS document also attributes the underlying data-sharing requirement to the model’s maker: “In order to use Claude Fable 5, as required by Anthropic, you must opt in to sharing retained traffic with Anthropic.”
That same passage adds one more detail that lines up with Anthropic’s own language about logging human access: the data is retained, AWS writes, “for abuse detection and potential human review.” Anthropic’s phrasing is close, logging “all human access to the data.” Both sides describe the possibility of a person looking at retained content. Neither one says every piece of retained content gets read by a person.
The judgment that holds up
“Zero retention” has no single answer that carries across channels and contracts. For the same model, Anthropic direct and AWS Bedrock each come with their own conditions, and the answer depends on which road you take, which contract you sign and what the vendor’s documents say at the time. Whether the difference between Anthropic’s and AWS’s documents has been resolved is something neither has said, so this piece does not pick a winner.
The two documents, laid side by side, don’t answer which company protects privacy better. They point toward a narrower, more useful set of questions. When does a vendor’s safety team actually look at content, and what does that process require: approval, a log, a notice? Which document is that written in, and would you find out if it changed?
A separate piece on this site compares Anthropic and OpenAI on different ground: Anthropic’s and OpenAI’s evaluation network failures, not retention terms.
Sources
Primary documents: Anthropic, Claude Fable 5 and Claude Mythos 5 (June 9, 2026); Data retention practices for Covered Models (July 9, 2026); Models overview; API and data retention; Privacy Center retention page. Amazon Bedrock official data retention and data protection documentation.
Frequently asked questions
What does Anthropic’s retention rule for Claude Fable 5 and Claude Mythos 5 actually require?
Per Anthropic’s Covered Models documentation, effective June 9, 2026, an organization that had configured zero data retention loses that option for these two models. Traffic is held for 30 days for safety review and is not used to train new models. Anthropic’s own text leaves itself an out: deletion happens after 30 days in almost all cases, not all of them. Mythos 5 itself is invitation-only and not generally available; the generally available model is Fable 5, which carries the same rule; Mythos 5 is limited to approved invitees. The rule only bites organizations that had zero retention configured in the first place, for the two models discussed here; other organizations’ retention settings are unchanged. AWS’s Bedrock abuse-detection page lists Fable 5, but not Mythos 5, among models for which eligible customers may request full ZDR through their AWS account team, and the article sets the two documents side by side. A mismatched configuration returns an API error, which is a rejected request, not evidence that the content was retained.