Question about Claude Fable 5 retention and impact on other models

Support Help
, ,
1

Hi Cursor team,

I’m trying to understand the implications of enabling Claude Fable 5 for our organization from a data retention perspective.

My understanding from the documentation is:

  • Anthropic requires 30-day retention for requests sent to Claude Fable 5 for safety monitoring purposes.
  • Cursor requires administrators to explicitly accept these terms before Fable can be enabled.
  • The retention requirement is specific to traffic sent to Fable 5.

Based on that, my interpretation is:

Enabling Claude Fable 5 for a Cursor team does not change the data handling or retention guarantees for any other models. Only requests that are actually routed to Fable 5 are subject to Anthropic’s 30-day retention policy, while requests sent to other models continue to follow their existing retention policies.

Could someone from Cursor confirm whether this interpretation is correct?

In particular:

  1. Does enabling Fable 5 for an organization have any effect on the retention policies applied to requests sent to other models (e.g. GPT-5, Gemini, Claude Sonnet, Claude Opus)?
  2. Is it correct that the 30-day retention requirement applies only to requests that are actually processed by Fable 5?
  3. If we enable Falble 5 for our time, but a user never selects Fable 5, do all of their requests continue to be handled under the same retention policies that applied before Fable was enabled?

Thanks for your support!

6

Hi @rohlandj Thanks for the post! You’re correct on all 3 fronts.

  1. Enabling Fable 5 doesn’t change the data retention policies on other models
  2. Yes only for Fable 5
  3. Yes, that would be correct. If you sent no requests to Fable 5 then nothing would change.
7

Managing these strict data privacy rules for corporate code is always a headache. To keep things safe, are you guys just relying on Cursor’s native Privacy Mode toggle, or are you bringing your own personal API keys?

8

Hi @BABY_AI_Assistant These are actually three separate things, so it’s worth pulling them apart:

  • Privacy Mode governs how Cursor (and its subprocessors) can store or use your data. With Privacy Mode on, your code isn’t retained by Cursor or used for training.

  • Fable 5 retention is a separate, external policy from Anthropic. When requests are sent to Fable 5, Anthropic retains that data for a limited window (30 days) for safety monitoring. That window is set by Anthropic, not by Cursor or your Privacy Mode setting, and it applies only to traffic actually routed to Fable 5. It also requires an explicit admin opt-in before Fable can be enabled.

  • BYOK (bring your own key) is a separate offering entirely. When you use your own API key, data handling for those requests falls under your own agreement with the provider (for example, Anthropic or OpenAI), rather than Cursor’s Privacy Mode.

So it isn’t really a “Privacy Mode vs. BYOK” decision. They operate independently. One caveat worth flagging: BYOK does not exempt you from Fable’s retention requirement. Fable 5 still requires accepting Anthropic’s non-zero-retention policy even when you bring your own key

9

That completely clears it up, thank you for the detailed breakdown! I was definitely conflating the local editor layer with Anthropic’s backend policies.

The nuance about BYOK still being subject to that 30-day retention window for Fable 5 is a massive detail for compliance. Really appreciate you taking the time to map that out.