I have previously analyzed the privacy policies of quite a few AI tools. Recently, Claude introduced the model called Fable, which is an AI model with substantially greater capabilities than prior models. Fable is based on Mythos, which you may recall Anthropic initially restricted to a small group of approved partners due to the potential chaos it could have caused for cybersecurity among other issues.
The Major Difference is the Data Retention Period
The critical difference as far as privacy and terms of service is that for those using a Fable model, regardless of what level of Claude you are using, Anthropic retains the prompt and resulting data for 30 days. Previously, certain levels of the tool had the potential for zero data retention (ZDR) based on the contract and your settings. With Fable, this goes away. This is the case even if you are using the enterprise level of the tool. In other cases, the data was already retained for at least 30 days, depending on the level of Claude you purchased, so for those users, this change means little.
Fable Has Limits Prior Models Did Not
Fable 5 also has hard limits built in that other models did not. If you ask Claude Fable something in a restricted area, like certain cybersecurity or biology questions, Fable won’t answer. Instead, Anthropic hands your question to an older model, Claude Opus 4.8, which answers instead. You may not even notice the switch.
One point is currently unclear: Opus 4.8 is normally a model that enterprise customers can use with zero data retention, meaning Anthropic stores nothing. But Anthropic has not made clear what happens to retention when Fable hands a question off to Opus. The policy says retention applies to Mythos-class models on every platform, which suggests the entire Fable session is retained regardless of which model answers. But Anthropic has not addressed this scenario directly. Until it does, the cautious assumption is that everything you type into a Fable session gets retained for 30 days.
Think of a law firm where one partner has strict confidentiality rules and another has looser ones. You call the partner with looser terms and tell him everything about your matter. He then hands your case to the partner with strict terms. The strict partner’s rules don’t reach back and protect what you already said. Your information is governed by the rules in place when you shared it. Fable appears to work the same way. You enter your prompt through Fable, so Fable’s retention rules attach at that moment, even if Opus 4.8 quietly does the answering.
The Current Takeaway
First, for much of your work, Opus may suffice and Fable may not be necessary. Second, remember that Fable is not a ZDR model. Don’t use it if you have concerns about data retention or your firm has AI use policies against allowing data to be retained by outside tools.