Astra for Law forces firms into governance per matter

You must now treat Astra for Law as regulated legal infrastructure, not as general-purpose ChatGPT. Make explicit choices per matter about which functions you enable, which client data enters the system, and how every output is logged, reviewed and attributed to a responsible lawyer.

An analysis of 20 September 2026 of how Astra for Law is deployed under bar ethics argues that the tool shifts the question for firms from whether they use AI to how they control this specific, legally tuned tier. OpenAI announced Astra for Law as GPT-6 tuned to legal work, coupled to a US Legal Search Index and instructions for legal analysis and drafting. In our assessment this represents a step change in the specificity of the tooling available to legal practice, which means the burden of demonstrable control now falls entirely on how you deploy it per matter.

What makes Astra for Law different from general ChatGPT?

Astra for Law appears as a separately labelled tier—"GPT-6 Astra Law" in the model picker and gpt-6-astra-law in the API—not as a hidden setting or configuration. It is a legally tuned version with its own search index and bespoke instructions, designed to support lawyer expertise rather than replace it. OpenAI names governance, privacy and ethics as design premises, signalling that this tier is intended for controlled, authorised deployment. That specificity matters: it means you cannot treat Astra as you might treat general-purpose AI tools. The tooling has already been made specific to legal work; your responsibility is to make it specific to each matter.

Which ethical rules now apply?

Bar ethics rules on confidentiality and responsibility for non-lawyer assistants apply directly. Rule 1.6 on confidentiality and Rule 5.3 on responsibility for non-lawyer assistants establish that you must treat Astra as a non-lawyer technological assistant requiring supervision. This is not new law; it is existing law applied to a new tool. What changes is the scale: because more work can run through AI, the need for demonstrable supervision grows. Every output must be reviewed by a responsible lawyer before it leaves the firm. That review must be recorded and traceable to the matter.

What controls must you demonstrate per matter?

You must record the following choices for each matter:

1. Identify the responsible lawyer — name the lawyer who will review all outputs from Astra before they are used or disclosed.
2. Document which functions are enabled — specify which Astra capabilities (intake, drafting, review, research) the matter will use.
3. Specify which client data enters the system — record what information about the client, the matter or the opposing party will be processed by Astra.
4. Log every output and its review — capture what Astra generated, when, for which purpose, and which lawyer reviewed it and when.
5. Record the lawful basis — document why this matter and this data flow through this tool under your engagement terms and applicable law.

Without this registration you cannot demonstrate afterwards how Astra influenced a matter or that supervision occurred. Bounding functions per matter aligns with the broader requirement to limit and log AI use by lawyers and to show control over each workflow.

What practical steps should you take now?

First, use Astra only on non-training accounts—Business, Enterprise or API tier—so that your interactions are not used to retrain the model. Second, establish a logging layer that captures every interaction: what was sent to Astra, what it returned, when a lawyer reviewed it, and what decision the lawyer made. Third, tie every step to a responsible lawyer and a checkpoint. Fourth, recognise that Astra for Law is currently available only on a limited basis via Trusted Access and is aimed at US law; do not assume it will work the same way in other jurisdictions or that pricing and availability will remain unchanged.

For legal-tech vendors integrating Astra into products, the requirement is the same: make visible which parts of an automated workflow are powered by Astra and under which controls. Astra for Law delivers a powerful but fallible legal tier. It belongs where the index, the controls and human supervision are demonstrable per workflow.

Tooling can capture what happened and when; it cannot decide whether supervision was adequate or whether the lawyer's review was genuine. That remains your professional judgement, and it must be recorded in a way that a regulator or a court can later examine.