Australian law firm Gilbert + Tobin is scaling ChatGPT Enterprise and Codex across the entire firm, according to a case study published by OpenAI on 1 September S¹. The rollout pairs CEO-led commitment with what OpenAI describes as "rigorous governance" and "human accountability" S¹. For a profession that bills by the six-minute increment and treats risk like a loaded weapon, that is a striking move. The question is what a law firm is willing to let an AI touch, and what it will not.

My read: This is the first full-firm AI deployment I have seen from an Australian law firm at this scale, and the detail that matters is the line G+T has drawn. They are not using AI to deliver legal advice. They are using it for the operational work that supports advice P². That is a smart boundary, but I would want to hear it from G+T directly rather than from OpenAI's own case study page. The vendor is not a neutral observer here.

Where the line is drawn

The case study, published on OpenAI's website, frames generative AI as a tool for improving the quality of behind-the-scenes operational work rather than the legal advice itself P². The firm aimed to equip all operations teams with access to the tools P². That means the AI is handling the scaffolding: document review pipelines, research triage, drafting support, the repetitive knowledge work that surrounds a lawyer's core judgement.

This matters because the biggest fear in legal AI is hallucinated citations, invented case law, a model confidently citing a precedent that does not exist. A firm that keeps AI away from the final advice product, and focuses it on operational workflows, is managing that risk by design.

What "governance" means here, and what it does not

OpenAI's case study credits G+T with three things: CEO-led commitment, rigorous governance, and human accountability S¹. Those are the right words. They are also the only words we have. The case study does not publish specific governance policies, adoption metrics, or efficiency gains. It does not say which practice groups are covered, or whether the rollout extends across all jurisdictions where G+T operates. It does not describe what "human accountability" looks like in practice, whether that is a lawyer signing off on every AI output, a review committee, or something lighter.

What we can verify is the toolset. ChatGPT Enterprise is OpenAI's business tier, which keeps customer data out of model training. Codex is OpenAI's coding tool. The combination suggests G+T is using AI for both text-based operational work and code, though the case study does not confirm what Codex is being used for specifically.

What to do about it

The takeaway for any professional services firm watching this: the boundary matters more than the brand of AI. G+T is deploying AI where the cost of error is operational, not legal. A drafting error in an internal research memo can be caught and corrected. A hallucinated citation in court cannot.

Consider a mid-sized conveyancing practice in Sydney. The firm could use an AI tool to triage incoming contract review requests, summarise title documents, and flag unusual clauses for a lawyer's attention. The lawyer still reviews, still signs off, still carries the professional liability. The AI compresses the reading time from two hours to twenty minutes. That is the operational layer G+T is targeting, and it is where most professional services firms should start.

One practical thing to check this week: review your firm's existing AI policy, if you have one, and ask whether it distinguishes between client-facing advice and internal operational work. If it does not, that gap is where the risk lives.

What we don't know yet

The evidence here is a single vendor-published case study. We have no independent statement from Gilbert + Tobin confirming the scope, no metrics on adoption or efficiency, and no detail on what "rigorous governance" actually means in practice S¹. The case study sits on OpenAI's website, which means it is promotional by nature. We also do not know whether Codex is being used for internal software development, legal coding work, or something else entirely.

The deeper question is whether this model holds. Law firms that draw a clean line between AI-assisted operations and human-delivered advice are betting that the line stays clean, that the tools do not creep toward the advice itself as they improve.

The next signal: whether Gilbert + Tobin independently confirms the deployment scope in its own communications before November 2026. If the firm stays silent, the OpenAI case study remains a vendor-published claim, not an audited deployment. We will check this claim against whatever G+T says next.

If you found this useful, subscribe for the next one. We follow the evidence, not the press release.

Sources: S1 — How law firm Gilbert + Tobin governs and scales AI with OpenAI · P2 — How law firm Gilbert + Tobin governs and scales AI with OpenAI · P3 — GitHub - SalesforceAIResearch/enterprise-deep-research at 40a8cc37c53f · P4 — openai/GABRIEL · P5 — Tobin South

More from Not A Tech Guy

Generated from an audited evidence pack with primary-source research. Social-media items are discussion signals, not verified facts. Nothing here is financial, legal or medical advice.