Claude Fable 5 Is Free Until June 22. Here's the Line You Can't Cross With Client Files.

June 13, 20266 min readBy The Crossing Report

Anthropic released Claude Fable 5 on June 9, 2026 — the most capable Claude model available on a standard Pro, Max, or Team subscription. Through June 22, it is free at your current plan level. Accounting firms, law firms, and consultancies have nine days to test the most powerful Claude model yet at no extra cost.

Here is the problem: there is a clause in Anthropic's current Fable 5 policy that most professional services firm owners have not read. And if they use Fable 5 the wrong way for the next nine days, they may hand client data to a retention window they did not agree to.

This is not a reason to skip Fable 5. It is a reason to understand the line before you cross it.


What Makes Fable 5 Different

Claude Fable 5 is Anthropic's "Mythos-class capability made safe for general use." In practical terms: it can handle multi-document tasks — reviewing 50 contracts, running a multi-state tax analysis, cross-referencing regulatory guidance across jurisdictions — and return a structured, review-ready result rather than answering one question at a time.

The 1-million-token context window is the spec that matters most for professional services work. That is roughly 750,000 words of input — the equivalent of four average-length novels, or a complete M&A due diligence document room. You can ask Fable 5 to review an entire contract lifecycle, a full client tax file's worth of publicly available data, or a research corpus without hitting the context ceiling that forces mid-task pivots in smaller models.

The capability is real. The free window through June 22 makes this the right moment to test it. But the way you test it matters.


The Retention Clause (What Firms Are Missing)

Anthropic is retaining inputs and outputs submitted to Claude Fable 5 for up to 30 days. This is standard practice for new frontier models under active safety monitoring. Inputs are stored in the US and deleted after the 30-day window.

For a professional services firm, that 30-day retention window is the governing factor for what you can safely submit.

Here is the practical consequence:

Anything you type into Claude Fable 5 through Claude.ai — including Claude Pro, Max, Team, or Enterprise accessed via the web interface — is retained by Anthropic for up to 30 days.

That means:

  • A client's tax return uploaded for analysis
  • A litigation strategy memo submitted for review
  • A financial statement with client-identifying information
  • A contract with confidential terms
  • Any engagement communication referencing a named client

All of it sits in Anthropic's retention window for 30 days, regardless of what you agreed to in your engagement letter with the client.

Microsoft reportedly restricted its own employees from using Fable 5 on internal work for exactly this reason: their internal governance standards require that employee work product not be retained by third-party AI providers without enterprise-level data processing agreements. If Microsoft's governance team reached that conclusion, professional services firm owners should reach the same one about client data.


The Three Uses That Are Off the Table

Client files, period. Tax returns, financial statements, engagement-specific documents with client-identifying information — none of these belong in a Claude Fable 5 session through Claude.ai until you have a zero-retention API agreement in place.

Legal briefs and confidential client communications. Attorney-client privilege does not transfer automatically. Submitting privileged communications to a third-party AI tool for review may create a privilege waiver risk, independent of the retention question. Fable 5 does not change this analysis — but its retention clause makes the risk more concrete.

Confidential advisory work with named client context. If the input contains the client's name, industry, financial position, or other identifying details alongside the work you are reviewing, it belongs outside the retention window.


What You Can Safely Test Before June 22

You have nine days. Here is what legitimate testing looks like with the retention clause in place:

Template development. Use Fable 5 to draft or improve your firm's internal templates — engagement letters, proposals, information requests, client onboarding checklists — with all client details removed or replaced with placeholder text. Fable 5's multi-document reasoning makes it exceptionally good at pattern-matching across template structures.

Internal policy drafts. AI use policy, data handling policy, conflict-of-interest procedures, staff training materials. None of these require client data. Fable 5 can draft a legally coherent AI use policy for a 15-person law firm in 10 minutes. That is a legitimate test of its capability that does not involve a retention risk.

Research on publicly available material. Ask Fable 5 to analyze a published regulatory update, synthesize a collection of court decisions, or identify patterns in publicly available industry data. The 1-million-token context window is transformative here — you can feed it an entire year of IRS guidance updates and ask it to flag the five most relevant to your practice area.

Summarization of published reports. The TR 2026 AI in Professional Services report, the AICPA Profession Ready initiative, state bar guidance on AI — all of these are public documents. Use Fable 5 to synthesize multiple published sources into a briefing for your team.

The test is simple: if Anthropic read this input tomorrow, would there be any client relationship or professional obligation that required it to stay private? If yes, it does not go into Fable 5 through Claude.ai.


The Path to Client-Safe Fable 5 Use

If you want to use Fable 5 for client work, two paths exist:

Claude API with a zero-retention enterprise agreement. Anthropic offers data processing agreements (DPAs) to enterprise clients that exclude inputs from the retention window entirely. Anthropic does not store what you submit. This requires a direct enterprise contract — not a Team or Pro subscription. For a 10-person firm, this is achievable but requires an enterprise negotiation, not a credit card.

Use Fable 5 through a vendor with enterprise data agreements in place. Tools like Harvey, Karbon Kai, and HubSync are built on Claude via API with their own enterprise data governance terms. When you use Harvey for contract review, Anthropic's consumer retention policy does not apply — Harvey's own data agreement with Anthropic, and Harvey's data agreement with your firm, govern what happens to the input. The same applies to any professional tool that has negotiated a zero-retention API arrangement.

If you are using Fable 5 through Claude.ai on a Pro or Team plan — the default subscription path — neither of those protections exists. The 30-day retention window applies.


One Action Before June 22

Test Claude Fable 5 on something that does not involve client data. Pick one internal document your firm writes repeatedly — a client proposal, an onboarding checklist, an internal research memo format — and ask Fable 5 to improve it. Use placeholder client names. Submit only publicly available information as context.

Spend 30 minutes with the model. See what it can actually do. You will have a concrete sense of the capability upgrade before the free window closes — and you will not have put a client relationship at risk to get it.

After June 22, Fable 5 requires usage credits. Before then, the test is free. Do the test. Skip the client files.

Frequently Asked Questions

Is Claude Fable 5 safe to use for client files at a law firm or accounting firm?

Not by default. Anthropic retains inputs and outputs submitted to Claude Fable 5 for up to 30 days for safety monitoring, processed in the US. This means client-identifying information, privileged communications, tax returns, financial statements, and confidential advisory work submitted to Fable 5 through Claude.ai (Pro/Team/Max) falls within the retention window. For client-safe use, you need either a custom API deployment with retention disabled (Claude API + zero-retention agreement) or an enterprise contract that explicitly excludes your data from retention. If you do not have one of those in place, do not use Fable 5 for client files.

What can professional services firms safely test with Claude Fable 5 before June 22?

Safe test cases for the free window include: internal document templates (engagement letters, proposals) with all client details removed or replaced with placeholder text; internal policy drafts (AI use policy, data handling policy, HR procedures); research tasks using only publicly available information; training materials for staff; summarization of published reports or regulatory guidance. The key test: would you be comfortable if Anthropic saw this input for the next 30 days? If no, it belongs in a zero-retention API deployment, not a Claude.ai subscription session.

What is Anthropic's data retention policy for Claude Fable 5?

For Claude.ai plans (Pro, Max, Team, Enterprise accessed via the web interface), Anthropic retains inputs and outputs submitted to Fable 5 for up to 30 days for safety review, with data processed in the US. After 30 days, they delete it. This policy applies because Fable 5 is a new model under active safety monitoring. For Claude API access with a custom enterprise agreement, firms can negotiate zero-data-retention terms that prevent Anthropic from storing inputs at all. The retention policy is the reason Microsoft reportedly restricted its own employees from using Fable 5 on internal work.

Why did Microsoft restrict employees from using Claude Fable 5?

According to reporting cited in Legal IT Insider (June 9, 2026), Microsoft restricted its own employees from using Claude Fable 5 on internal work because of Anthropic's 30-day input/output retention policy. Microsoft's internal governance standards require that employee work product not be retained by third-party AI providers without enterprise-level data processing agreements. This is not unique to Fable 5 — the same logic applies to any AI model where the provider retains inputs. It is a useful signal for professional services firms: if your AI governance policy mirrors enterprise software standards, the same restriction likely applies to your client-facing work.

How do I use Claude Fable 5 for client work without the retention issue?

Two paths: (1) Claude API with zero-retention enterprise agreement — Anthropic offers this to enterprise clients who negotiate a data processing agreement (DPA) that includes zero-retention terms. Inputs are not stored. This requires a direct Anthropic enterprise contract. (2) Use Fable 5 through a vendor that has negotiated enterprise terms — tools like Harvey, Karbon Kai, and HubSync build on Claude via API with their own enterprise data agreements. In those tools, Anthropic's consumer retention policy does not apply; the vendor's data governance terms do. Always check the vendor DPA before using client data in any AI tool.

Get the weekly briefing

AI adoption intelligence for accounting, law, and consulting firms. Free to start.

Related Reading

This is the kind of intelligence premium subscribers get every week.

Deep analysis, cross-sector patterns, and the frameworks that help professional services firms make the crossing.