OpenAI's Astra for Law Is Gated to Elite Firms. That Gap Is a Sellable 'Vertical AI Readiness' Audit for Everyone Else.

OpenAI's Astra for Law is not just a smarter model with a law degree. It's a vertical AI product with a locked door — and that locked door is where the business is.

What actually happened

OpenAI tested Astra for Law against "200 U.S. legal research questions" and reported a "54.0%" pass rate on overall correctness, versus "38.7% for GPT-6 Astra with web search alone" — OpenAI frames that as "a 40% relative improvement." On case-law-focused questions, it found "24% more" reference cases than the web-search baseline, and pulled "up to 54% more relevant passages from correct court opinions." The plugin layer already has named specialist partners — Thomson Reuters, Harvey, Legora, and iManage — and the offering includes "Zero Data Retention (ZDR) on API for eligible firms," a specific compliance commitment rather than a vague privacy promise.

None of that is available broadly yet. It's Trusted Access to a short, elite list, with the API "coming soon" and no public date attached.

What this exposes

• A real vertical AI product now exists with a named performance gap — 54.0% vs 38.7% on a 200-question benchmark is a specific, checkable number, not marketing language, which makes it a real bar for anyone building competing or complementary workflows to measure against
• The plugin ecosystem is being built in public, right now — 26 partner plugins plus 47 community plugins means the integration map already exists; firms just need someone to translate it into their own stack
• Access is the bottleneck, not capability — Trusted Access to four named elite firms while "API access" is only "coming soon" means every other firm and vendor has a real waiting period to use productively or waste
• Compliance claims need firm-specific verification — "Zero Data Retention... for eligible firms" is a conditional claim ("eligible") that a firm's own counsel will want re-checked against their specific confidentiality obligations, not taken at face value

The business idea

Any mid-size firm, boutique practice, or legal tech vendor that isn't Latham & Watkins is sitting in the same spot right now: a real product exists, the benchmark numbers are public, and there's no self-serve way in yet. That gap is a fixed-scope readiness engagement:

• Map the client's existing case management and document stack against the 26 already-shipped partner plugins, so they know precisely what activates the moment access opens
• Replicate OpenAI's own benchmark methodology — a fixed question set, pass-rate and reference-case scoring — against the client's actual past matters, since a general 54% number says nothing about a niche practice area
• Help vendors outside the initial 26 partners build and ship a community plugin before the ecosystem consolidates around whoever moved first
• Stand up an interim workflow today with ChatGPT for Word and the community plugin layer, so the client is capturing value and tuning instructions before their own invite lands
• Sell the ZDR and data-retention compliance check as a recurring retainer line, since "eligible" is a condition that will need re-verification every time OpenAI adjusts the offering

Why this works now

Law firms are risk-averse by training, not laziness — confidentiality obligations mean they will not touch a new AI tool without someone mapping the compliance and integration surface first. OpenAI just proved the category works with real numbers and real plugin partners, which removes the "does this even work" objection. What's left is entirely execution: stack mapping, benchmark replication, and compliance verification — work that doesn't require an OpenAI partnership to sell, only the ability to read what OpenAI already published and apply it to one firm at a time.

Bottom line

OpenAI didn't open Astra for Law to the market. It opened it to four named firms and a plugin ecosystem, with everyone else told to wait for an API that's merely "coming soon." That waiting list is a client list, and the audit work needed to be ready for it is sellable today.

Sources:

https://openai.com/index/astra-for-law/

https://www.artificiallawyer.com/2026/09/18/openai-launches-astra-for-law/