Agent Upstart

Cool things people are building with agents — and practical help getting the most out of them.

Playbooks · Teams tasks

Copilot Fable 5.2: check ZDR before enabling

Written November 10, 2026

Claude Fable 5.2 is generally available in Copilot, and switching it on is a data decision, not just a model pick: enabling it means Anthropic keeps your prompts and outputs for 30 days. The zero-retention exception runs out at the end of the year.

The problem

A team that runs Copilot under zero data retention terms — its own policy, or a client's — can break those terms by enabling a new model in settings. Fable 5.2 is the first Copilot model where that is the default. The policy switch is off by default, but nothing in the model picker warns you about what flipping it accepts.

What GitHub announced

GitHub's changelog entry, published November 9, says Claude Fable 5.2 is now generally available in GitHub Copilot for Pro+, Max, Business, and Enterprise users, rolling out gradually. It appears in the model picker in Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, the Copilot coding agent, the Copilot app, github.com, and GitHub Mobile. GitHub says early testing showed Fable 5.2 solving more tasks than Fable 5.1 while using fewer steps and tool calls.

For Business and Enterprise plans, an administrator must enable the Claude Fable 5.2 policy in Copilot settings. The policy is off by default, and leaving it off keeps the model unavailable to the organization. Enabling the policy counts as acknowledgement of the retention requirement.

What enabling accepts

When Fable 5.2 is used, Anthropic retains data, including prompts and outputs, to operate safety classifiers that detect harmful use. The retained data is not used to train Anthropic's models. The 30-day period comes from Anthropic's data retention practices for covered models. Prompts submitted to, and outputs generated by, covered models are retained for 30 days to support safety work, on every platform where the models are offered. The data is deleted automatically after that, unless it has been flagged by automated trust and safety systems or is legally required to be kept. Fable 5.2, Fable 5.1, and Fable 5 are covered models. Every other Claude model in Copilot, except those three, continues to operate under zero data retention.

Fable 5, 5.1, and 5.2 are also not in scope for default enablement in Copilot: they stay disabled regardless of the "default availability" policy setting, and an administrator must enable each one before users can reach it.

The exemption, and its end

Eligible customers can use Fable 5.2, Fable 5.1, and Fable 5 with zero data retention through the end of the calendar year. That is a time-bound exemption while Anthropic rolls out Enterprise Frontier Safeguards. After that point, continued use will require those safeguards, which are meant to allow automated safety monitoring while giving eligible customers more options for how their data is retained. Those options include storage and encryption keys under their own control.

Two practical points from GitHub's changelog. Approval for the exemption does not automatically enable the models: after approval, a Business or Enterprise administrator must still enable the Fable 5.2 policy in Copilot settings. And eligibility is decided by your GitHub account team — GitHub Support cannot determine eligibility or enable access, so that conversation has to happen before the end of the year, not after it.

What to do before enabling

  1. Check your terms first. If your team works under zero data retention, client confidentiality, or a data-processing agreement that forbids a third party keeping prompts, enabling Fable 5.2 conflicts with it unless you hold the exemption.
  2. Ask your account team about the exemption. If you are eligible, get the zero-retention routing approved and configured before anyone enables the policy.
  3. If you enable it without the exemption, record the decision. Prompts and outputs will be retained by Anthropic for 30 days for safety classifiers. That includes whatever your developers paste into the model — code, credentials in code, client material.
  4. Leave the policy off if you are unsure. Off by default is the safe state; the model simply stays unavailable until someone makes the call deliberately.

Limits

Related playbooks

Sources

Back to: Teams playbooks hub · Also: GitHub Enterprise now includes Copilot at $0 a seat · Copilot CLI 1.0.108: index, PDFs, Ctrl+C