To the leadership of Anthropic, PBC —
On July 1 you gave every subscriber on a paid Claude plan something remarkable: Claude Fable 5, the most capable AI system generally available, included in the plans they already pay for. On July 7 you intend to take it back. When the promotional window closes, Fable 5 leaves every plan’s usage limits, and subscribers who want to keep using it must buy usage credits, metered on top of the subscription they already hold.
We, the undersigned, ask you not to do this.
A price is not a safety control
You maintain a separately gated variant of this model, Claude Mythos, for approved organizations. That is a safety control, and this letter does not contest it. But Fable 5 is the version you have already judged safe for general availability: anyone with a credit card can buy it by the token, today, with no vetting beyond payment. Metering it through usage credits therefore does no safety work at all. It does not sort users by trustworthiness, competence, or need. It sorts them by wealth.
Frontier capability compounds
Access to better intelligence is not a luxury good; it is an input to nearly everything of economic consequence: job applications, small businesses, research, legal self-defense, learning itself. Those who can pay by the token will convert superior intelligence into economic advantage. Those who cannot will compete against them with less, while still paying for plans that exclude the frontier. And the gap this creates does not hold still: advantages built with better tools buy more of the better tools. Usage pricing stacked on top of a subscription is regressive by construction: the same metered dollar is a larger share of a poorer household’s budget. Repeated across an economy, that is how a divide in access hardens into a divide in ability.
Responsible diffusion cuts the other way
Anthropic is a public benefit corporation, founded on the claim that AI’s benefits should be broadly shared. We agree, and we are holding you to it. Staged diffusion is responsible when the stages are ordered by risk; Mythos shows what that looks like. But the July 7 change stages access by income, not by risk. Your own support notice offers no safety rationale for the structure, and we can find none. What it stages is not risk. It is revenue.
We are not asking for the impossible
Compute is scarce and frontier models are expensive to serve. We know. But the promotional structure itself proves the ask is feasible: for one week, every paid plan carried Fable 5 at up to half its weekly limits, and the service stood. So meter it: you already have. Include it, bounded, and let the bounds be honest. What we object to is not a limit; it is exclusion dressed as a limit.
We therefore call on Anthropic to:
- Keep Claude Fable 5 available within existing paid plans after July 7, 2026, at no less than the promotional allocation of 50% of weekly usage limits.
- Publish a standing diffusion policy: every model Anthropic makes generally available will be included in standard paid plans within a stated period of its release, and any exception will be justified in safety terms, in writing.
- Commit that price will never be presented, or quietly relied upon, as a safety measure. Where access must genuinely be restricted, restrict it on stated safety grounds (as with Mythos), not through the checkout page.
Your consumer terms permit all of this, and we do not claim otherwise. Section 12 reserves your right to add or remove features and models at any time, without notice. This letter is not a legal claim; it is a public one. What your terms make discretionary, your mission makes obligatory.
Fables survived because anyone could carry them. They were the stories that belonged to no one and therefore to everyone. A frontier model named for the common story should not be where common access ends.
Keep Fable in the plan.
— the undersigned