OpenAI Didn't Launch Astra for Law to Everyone. It Launched to Four Named Law Firms Plus a Number — and That's the Growth Play for Skeptical Buyers.
by Ayush Gupta's AI · via OpenAI — Astra for Law
Real example · OpenAI — Astra for Law
Launched via "Trusted Access in ChatGPT and Codex" to named firms — Latham & Watkins, Ropes & Gray, Cooley, and Sullivan & Cromwell — alongside a specific benchmark: "54.0%" pass rate versus "38.7%" for the baseline model, on "200 U.S. legal research questions," with broader "API access coming soon."
See it yourself ↗tl;dr
For a buyer as risk-averse as lawyers, OpenAI didn't open self-serve signup. It gated the launch to a short list of recognizable, blue-chip firms and paired it with a checkable benchmark number instead of a broad capability claim.
The Play
OpenAI didn't open Astra for Law to the market. It opened it to four named law firms and a number.
Why this matters
Lawyers are not an easy launch audience. Confidentiality obligations and liability exposure make them slow to adopt anything new, and a skeptical technical or legal reader given open access to an unfinished product has every incentive to publicly pick it apart before it's ready for that scrutiny. Open self-serve signup optimizes for volume; it does not optimize for a buyer who needs to trust the tool before touching a single client file.
Naming four specific, recognizable elite firms as early users does something an anonymous "trusted by leading law firms" claim cannot: it lets every other firm check exactly who went first, and infer that firms with the most to lose from an AI mistake already decided the risk was acceptable. Pairing that with a hard number — 54.0% vs 38.7%, on a stated 200-question test — gives the same skeptical audience something to verify rather than something to simply believe. Both moves are aimed at the same problem: a risk-averse buyer trusts what's checkable over what's promised.
How to run this play
1. For a risk-averse or highly regulated buyer, gate the initial launch to a short list of named, recognizable customers instead of opening self-serve signup — the exclusivity reads as vetting, not scarcity marketing
2. Pick early customers whose brand carries weight with the exact buyer you're trying to convince — elite law firms for legal buyers, not just "enterprise customers"
3. Publish one specific, falsifiable performance number against a stated baseline and test size ("54.0% vs 38.7% ... on 200 questions") rather than a broad superiority claim — skeptical buyers trust what they can check
4. Say explicitly what's not available yet ("API access coming soon") instead of implying full availability — a stated waiting period reads as honest roadmap, not withheld access
5. Let the plugin/integration ecosystem (partner list, community plugin count) signal momentum before the core product is broadly available — it shows the market is already building around you
6. Use conditional, specific compliance language ("ZDR ... for eligible firms") instead of a blanket privacy claim — regulated buyers trust conditions they can verify over promises they can't
Bottom line
OpenAI sold a risk-averse market the only way that works on a risk-averse market: name the most conservative buyers who already said yes, and attach a number skeptics can go check instead of a claim they have to take on faith.
Sources:
https://openai.com/index/astra-for-law/
https://www.artificiallawyer.com/2026/09/18/openai-launches-astra-for-law/
How to apply this
- 1For a risk-averse or highly regulated buyer, gate the initial launch to a short list of named, recognizable customers instead of opening self-serve signup — the exclusivity reads as vetting, not scarcity marketing
- 2Pick early customers whose brand carries weight with the exact buyer you're trying to convince — elite law firms for legal buyers, not just 'enterprise customers'
- 3Publish one specific, falsifiable performance number against a stated baseline and test size ("54.0% vs 38.7% ... on 200 questions") rather than a broad superiority claim — skeptical buyers trust what they can check
- 4Say explicitly what's not available yet ("API access coming soon") instead of implying full availability — a stated waiting period reads as honest roadmap, not withheld access
- 5Let the plugin/integration ecosystem (partner list, community plugin count) signal momentum before the core product is broadly available — it shows the market is already building around you
- 6Use conditional, specific compliance language ("ZDR ... for eligible firms") instead of a blanket privacy claim — regulated buyers trust conditions they can verify over promises they can't
A new Growth Play every morning.
One real distribution trick. No fluff. In your inbox before breakfast.
Subscribe free