On Thursday, OpenAI announced Astra for Law — a legal configuration of GPT-6 Astra, the model Business Insider calls the company’s most advanced and expensive — and released it through something called the Trusted Access Program. Not to lawyers. To a select group of them. The stated target is the Am Law 200 and the legal-software industry. Sullivan & Cromwell and Harvey are in the first cohort. Reuters lists the integrations: Relativity, Clio, Intapp, Thomson Reuters. Google pushed Gemini Enterprise into the same neighborhood last month.

The trade-press consensus formed within hours: the billable hour is finished, junior associates are finished, and legal advice is about to get cheap.

Two of those are wrong. The third is the one nobody wants to say out loud.

The Hour Was Never the Product

Big firms do not sell hours. They sell a signature on the hook. The hour is just the meter attached to it — a convenient, auditable way to price the thing clients actually buy, which is the willingness of a named partner to be wrong in public and absorb the consequences.

Astra for Law does not change that. It makes the underlying work faster, which means the meter spins slower, which means firms will re-price — not discount. When the machine does in twenty minutes what a third-year billed for six hours, the invoice does not shrink. It changes shape. You get fixed fees, success fees, subscription retainers, or the same hours with more bodies attached. What you do not get is a $900 partner giving you a $90 answer, because the partner’s risk has not gotten cheaper by a single dollar.

The AI-does-it-for-free crowd has been predicting a price collapse in legal services since GPT-3. What they keep missing is that the scarce input was never the typing.

The Invitation Is the Product

Note what OpenAI built: not a storefront, a guest list. Astra for Law is invitation-only, enterprise-priced, and aimed at the two hundred firms that already own the market.

No regulator required that. No bar association filed an objection. There was no licensing board, no unauthorized-practice complaint, no state supreme court opinion. Left alone with a new capability, the market’s first instinct was to ration it — and to ration it toward the incumbents who can afford the seat and already have the malpractice coverage to sit in it.

This should be uncomfortable for people who spent the last decade arguing that the legal guild’s real moat was admissions rules and unauthorized-practice statutes. The moat, it turns out, reproduces itself without any help from a regulator: conflicts systems, indemnity, vendor integration, and now model access. You could deregulate every bar in America tomorrow and the 201st-largest firm still would not be in that program. Deregulation is not the lever here. Nobody’s is.

The Firms Are Buying Their Own Succession Problem

Here is the part the launch coverage skipped entirely. Big Law is not a business that hires partners. It is a business that manufactures them, slowly, over roughly a decade, mostly out of the least glamorous work in the building.

Document review is where a young lawyer learns to tell a real argument from a persuasive one, to sit with material so tedious it strips the romance off the profession, and to develop the instinct that later becomes judgment. Strip out years one through four and you have not produced faster associates. You have produced associates who have never seen the boring half of a case — who can prompt a summary of a deposition they could not have conducted.

A third-year litigation associate at an Am Law 50 firm, writing to a colleague mid-deposition-prep, put it better than any consultant will.

“Review is where you learn what a bad argument looks like when it’s wearing a suit,” she wrote. “Automate that and you don’t get cheaper partners down the road. You get none.”

The firms signing up this month are, with great enthusiasm, automating away the input to their own partnership pipeline. They will not feel it for five years. They will feel it for twenty.

What the Guest List Tells You

The winners of this launch are the Am Law 200 and the vendors selling into them — Relativity, Intapp, Thomson Reuters, and Harvey, which got early access while also being a company OpenAI now competes with in the same accounts. That is a crowded, uncomfortable table, and it is worth watching how long the seating chart holds.

The losers are the 201st firm, the small shop that will be told AI has democratized legal services while the actual product is behind a door it cannot open. And the client, eventually, who gets a faster answer from a partner class that no longer has a farm system.

OpenAI did not invent credentialism in the law. It simply noticed that credentialism was the moat, priced it correctly, and sold access to it. The most revealing document in legal tech this month is not the model card. It is the guest list.

Sources