Franklin AI News Brief

Astra for Law: Enterprise customers can request access

Key Takeaways

  • OpenAI updates Astra for Law on GPT-6.2 Astra and lets Enterprise customers request trusted access, while reporting qualified legal benchmark gains.
  • OpenAI has updated Astra for Law on GPT-6.2 Astra.
  • ChatGPT Enterprise customers can request access through the Trusted Access Program, including API access.

OpenAI announced an updated Astra for Law on November 17, 2026, and expanded who can request access. The configuration combines GPT-6.2 Astra with legal search and task-specific instructions.

Who can request Astra for Law?

The announcement says all ChatGPT Enterprise customers can apply through the Trusted Access Program, including for API access. A request is a route into that program; it should not be treated as an automatic approval or a self-service entitlement for every account.

What do the legal results show?

OpenAI reports a 20.8% full-credit rate on LAB-AA, compared with 15.0% for ordinary GPT-6.2 Astra. Full credit requires an assignment to meet all criteria without a material unsupported claim. This is a benchmark result, rather than evidence that a particular legal draft is correct.

Franklin analysis: a strict full-credit score helps explain why a generally useful draft may still need substantial review. A missing requirement and an unsupported assertion can each make the deliverable unsuitable even when much of the text reads well. A legal team should inspect the answer’s sources and the complete assignment, rather than evaluate it on fluency alone.

What should a legal team check next?

Start with a familiar closed matter or a comparable internal exercise where the reviewers know the relevant record. Define the research question, required jurisdictions and deliverable before examining the output. That makes it easier to distinguish a better search result from a better explanation.

Keep the access decision separate from the decision to rely on a result. Requesting the configuration establishes neither its fit for a firm’s information rules nor the accuracy of a particular citation. The practical next step is a reviewed evaluation of the team’s own work after access is granted.

Our read

Franklin AI Take

For an Enterprise legal team, the change makes an access request worth considering for a defined research workflow. The full-credit benchmark remains a limited evaluation. After approval, compare a familiar assignment against its source record and keep a qualified reviewer responsible for the resulting work.