Short answer: Anthropic’s EFS is a customer-controlled privacy and misuse-safeguard design, while GitHub says Fable 5.1 in Copilot retains data by default for safety classifiers unless an eligible, time-bound ZDR path applies.
The two 1 September 2026 announcements answer one enterprise-operations question: what must a buyer verify before enabling a frontier model in a workflow that has privacy, retention or audit requirements? The answer depends on the product surface, plan and approved controls.
The short answer for an enterprise buyer
Anthropic’s Enterprise Frontier Safeguards (EFS) and GitHub Copilot’s Claude Fable 5.1 announcement should be read together when a team is deciding whether to enable a frontier model in an enterprise workflow. The capability question is only half of the approval: the other half is where prompts and outputs are retained, who controls review, and which safety mechanism is available.
Anthropic says EFS combines zero-data-retention privacy with misuse safeguards and customer-controlled cloud storage. GitHub says Fable 5.1 is generally available in Copilot, but its safety classifiers require data retention by default. Eligible customers may use Fable 5.1 with ZDR through the end of 2026 under a time-bound exemption while EFS becomes available. That is a documented transition, not a promise that every account already has the exemption.
Keep the page distinct from the Fable 5 versus Sonnet comparison and the OpenAI zero-data-retention context: those pages answer model-selection and neighboring privacy questions, while this page explains the current enterprise control path.
| EFS | Anthropic’s enterprise design combining ZDR, misuse safeguards and customer-controlled storage. |
|---|---|
| Fable 5.1 in Copilot | GitHub says generally available with gradual rollout across listed plans and surfaces. |
| Default retention | GitHub says Fable 5.1 retains prompts/outputs by default to operate Anthropic safety classifiers. |
| Temporary path | Eligible customers may use Fable 5/5.1 with ZDR through end-2026 under a time-bound exemption. |
| Approval | Account, plan, region, cloud and policy details must be verified before production use. |
What Enterprise Frontier Safeguards changes
Anthropic says EFS is being developed with more than 100 customers and major cloud partners. The design lets a customer use its own cloud storage, customer-managed keys, access controls and audit logs. Customers can control monitoring and data review, and Anthropic says automated monitoring does not require Anthropic human review.
Anthropic says EFS supports Claude Code, Claude Enterprise and Claude Platform as well as partner surfaces including Bedrock, AWS, Google Agent Platform and Microsoft Foundry. Broad availability is described as later in fall 2026 and the current announcement points eligible customers to a request form. Do not treat the announcement as proof that a specific tenant can turn it on today.
The commercial boundary is also explicit. Anthropic says EFS does not change model behavior, API pricing or rate limits, and Anthropic does not charge for the feature. A customer can still incur cloud-provider costs for storage, reads, writes and egress. Put those costs in the architecture estimate instead of calling the control free.
- Customer-owned storage and keys are a control choice, not an automatic compliance result.
- Access and audit configuration remain the customer’s responsibility.
- EFS rollout timing is later in fall 2026, not a fixed date in the announcement.
- Cloud-provider usage costs remain possible even when Anthropic does not charge for EFS.
What GitHub says about Fable 5.1
GitHub’s 1 September Changelog says Claude Fable 5.1 is generally available in Copilot and is rolling out across Copilot Pro+, Max, Business and Enterprise, with surfaces including the coding agent, GitHub.com, mobile, IDEs, CLI and the Copilot app. The page describes the rollout as gradual, so a plan name alone is not entitlement proof.
For enterprise governance, GitHub says Business and Enterprise administrators must enable the model policy. Fable 5.1 is off by default in that setting. GitHub also says the model requires data retention by default to operate Anthropic’s safety classifiers, and that retained prompts and outputs are not used to train Anthropic.
The same announcement describes a time-bound path for eligible customers: Fable 5 and Fable 5.1 can use ZDR through the end of 2026 under an exemption while EFS becomes available. The approval does not automatically enable the model. Record the account-team decision and the expiry path in the production approval.
| Plan | Pro+, Max, Business and Enterprise are listed; the live account still controls access. |
|---|---|
| Admin policy | Business and Enterprise administrators must enable the model policy; it is off by default. |
| Retention | Default retention is described as necessary for Anthropic safety classifiers. |
| Training | GitHub says retained prompts and outputs are not used to train Anthropic. |
| ZDR exemption | Eligible customers may use ZDR through end-2026; approval does not automatically enable Fable 5.1. |
Approval checklist for marketing operations
First identify the surface. A prompt sent through Copilot, Anthropic’s API, Bedrock or another partner may have different retention, logging and administrator controls. The model name is not a sufficient data-flow description.
Then ask four questions: what content enters the model, where prompts and outputs are stored, which safety classifiers or reviewers see them, and who can retrieve or delete the records. Add keys, roles, audit logs, region and retention period. If the workflow handles customer, employee or regulated data, route the decision through security and legal review.
Operationally, pair this with Claude Tag for Marketing Teams and Claude’s browser and files guide. The point is to map the workflow and permissions, not to assume that a model’s product page answers the organisation’s retention policy.
- Name the exact product surface and plan.
- List prompt/output data classes and retention period.
- Confirm customer cloud, key, access and audit ownership.
- Capture safety-classifier and human-review boundaries.
- Record entitlement, exemption approval and end-of-2026 follow-up.
What remains unknown
The announcements are detailed, but they do not provide a universal availability date, a tenant-specific entitlement result or a complete implementation recipe for every partner surface. They also do not turn an architecture into legal compliance. Those are deliberate handoff points for the buyer’s security, procurement and account teams.
The safest publication and implementation language is therefore conditional: Anthropic says, GitHub says, eligible customers may, and the account team must confirm. Do not write that Fable 5.1 is ZDR for everyone or that EFS is already available everywhere.
Recommended workflow
- Map the exact model surface, plan, region and data flow.
- Confirm the administrator policy, retention default and safety-classifier requirement.
- If pursuing ZDR, obtain and retain the eligible-customer exemption evidence.
- Evaluate EFS storage, keys, access, audit and cloud-provider cost ownership.
- Set a dated re-check for EFS availability and the end-2026 exemption boundary.
Related Digital Marketer guides
- OpenAI zero-data-retention context — A neighboring enterprise privacy decision.
- Claude Fable 5 versus Sonnet 5 — Existing model comparison owner.
- Claude computer-use and file workflow guide — Operational guardrails for browser, skills and file workflows.
- Claude Tag for Marketing Teams — Slack-based Claude workflows, memory and spend controls.
- Anthropic alignment and security changes — Earlier Anthropic security context and open questions.
- Claude Code Auto Mode — Permission and classifier context for agentic work.
Questions marketers are asking
Is Claude Fable 5.1 ZDR by default in GitHub Copilot?
No. GitHub says data retention is required by default for Anthropic safety classifiers. Eligible customers may have a temporary ZDR exemption through the end of 2026.
Does EFS change Claude pricing or rate limits?
Anthropic says EFS does not change model behavior, API pricing or rate limits and that Anthropic does not charge for EFS. Cloud-provider usage costs can still apply.
Does a plan name guarantee Fable 5.1 access?
No. GitHub describes gradual rollout and admin policy. Verify the live organisation, plan, region and enablement state.
Does customer-controlled cloud storage equal compliance?
No. It is a control in the architecture. Keys, roles, audit, retention, data classification and legal requirements still need review.
Sources and dates
Publication, event, and update dates are kept separate below. A source without a stated publication date is marked accordingly; access date is 2 September 2026.
- Anthropic: Enterprise Frontier Safeguards — published/event date: 2026-09-01; accessed 2 September 2026. Anthropic describes customer-controlled cloud storage, monitoring choices, rollout timing, supported surfaces and cloud-provider cost boundaries.
- GitHub Changelog: Claude Fable 5.1 generally available in Copilot — published/event date: 2026-09-01; accessed 2 September 2026. GitHub describes plans, surfaces, gradual rollout, admin policy and the default retention needed for Anthropic safety classifiers.
Editorial note: Re-check platform documentation immediately before implementation. Product rollouts, eligibility, pricing, policy, and measurement definitions can change after this draft.
