OpenAI dots for lawyers: useful workflows and firm boundaries

A legal team rarely needs another confident paragraph. It needs the right document, the current version, a reliable chronology and a clear account of what still needs a lawyer's attention. That is the lens through which we would assess OpenAI's dots announcement.

This is not the launch of a separately announced lawyer product. OpenAI has previewed specialist enterprise dots and named commercial contracting among its internal testing areas. The legal workflows below are Blue Canvas pilot suggestions, not a claim that dots can independently practise law or are approved for confidential client work.

What changes with an ongoing agent?

A one-off chat starts with the material someone gives it. An ongoing workflow could track an approved source over time, prepare updates and flag new questions. The potential value is continuity, but continuity also makes permissions and matter separation more important.

OpenAI says specialist dots will have their own identities and access to defined systems, starting with enterprise pilots. Read our enterprise dots overview for the distinction between that preview and the general rollout.

Start with a matter preparation pack

For an initial test, use a fictional matter or an approved, appropriately sanitised document set. Ask the agent to produce a document index, a dated chronology and a list of missing information. Every event in the chronology should point back to the document and location that supports it.

Give the reviewer deliberately awkward cases: two documents with similar names, an undated note, a later correction and conflicting accounts of the same event. The useful behaviour is to surface the disagreement, not silently select a convenient answer.

Keep the output as a working aid. The responsible lawyer verifies the underlying material before relying on it in advice, correspondence or a filing.

Other workflows worth testing

• Client meeting preparation: assemble approved background material and open questions, without contacting the client.
• Contract comparison: identify changed wording against a supplied version and flag departures from a firm-approved checklist. A lawyer interprets the effect.
• Internal knowledge maintenance: flag guidance that may need review when an approved source changes. The knowledge owner decides whether to update it.
• Administrative follow-up: prepare a list of missing documents and draft a reminder for review, without automatically sending it.

Whether any of these can be implemented depends on the product available to the firm, its systems and the permissions agreed. Begin with an output you can inspect before making persistent changes.

Keep professional judgement and consequential actions with people

Our proposed pilot boundary excludes autonomous legal advice, court submissions, settlement commitments, client-money activity and changes to critical deadlines. Do not make an agent's inferred date the sole source for a deadline. Do not accept a case citation simply because it looks plausible.

For research, require direct sources and verification by a qualified reviewer. If the agent cannot support a statement, it should flag the gap. The appropriate answer can be that the evidence is insufficient.

Review confidentiality before connecting a matter system

Ask the firm's relevant legal, information-security and data-protection owners to assess the actual proposed configuration. That review should cover authorised users, matter boundaries, retention, connected services, records of actions and how access is revoked. Product safeguards are not, by themselves, a determination that a workflow meets a firm's duties.

OpenAI's safety page notes that disconnecting an app stops new sharing through that connection, but information already learned can remain in a dot's context. That makes the offboarding and context-reset process worth testing, not assuming.

What would make the pilot worthwhile?

Compare the time to produce and verify a pack with the current process. Record omissions, unsupported assertions and corrections, not just drafting speed. A small, consistently accurate administrative workflow is more valuable than an ambitious demonstration that nobody can confidently sign off.

This is workflow guidance, not legal advice. Firms should assess their own professional obligations and client arrangements before using any new system with live matters.

Choose a useful first pilot

Blue Canvas can help you map a workflow, define its approval points and decide what to test before connecting business systems. Start with an AI audit or explore our AI implementation and automation support.

Book a free 15-minute consultation. Bring one recurring task, the systems involved and the decision you would never want an agent making alone.

Sources and scope

Based on OpenAI's dots announcement and safety, security and privacy explanation, published on 29 September 2026. Workflow designs below are Blue Canvas recommendations, not tested client results or promises of product capability. Blue Canvas is an independent consultancy; this article does not imply OpenAI partnership or access to its specialist enterprise pilots.