Summary
- At Australian law firm Gilbert+Tobin, 87% of active ChatGPT Enterprise accounts were actually being used as of June 2026
- Codex directly handles tasks like preparing audit reports for 300 entities and organizing 1,100 files, all under human review
What an 87% active-use rate actually means
According to a customer case study published by OpenAI, 87% of ChatGPT Enterprise accounts activated at major Australian law firm Gilbert+Tobin were actually being used as of June 2026. That's more than double the adoption rate typically seen with internal tools. What stands out is that this level of usage came from a firm focused on capital markets, M&A, and dispute resolution, spanning partners and non-technical staff alike.
Law firms are built around confidentiality and professional accountability. Handling client information and carrying responsibility for every judgment call naturally makes bringing in generative AI a cautious proposition. Rather than trying to replace legal advice itself with AI, Gilbert+Tobin chose to focus on raising the quality of the operational work that supports that advice. ChatGPT Enterprise was first rolled out to operations teams to validate real-world use cases in a controlled environment, then expanded in stages to marketing, business development, recruiting, finance, technology, business transformation teams, and parts of the legal team.
Building trust through leadership and governance
The available sources don't provide enough detail to support this section further. That kind of messaging helped establish AI use as a legitimate way of working across the firm.
That said, the firm didn't set mandatory usage targets. Instead, the business transformation team joined department-level meetings directly, demonstrating use cases tailored to marketing, finance, recruiting, and other functions. On top of that, internal videos, shared case studies, and challenges helped employees keep pace with new ChatGPT features.
Because confidentiality requirements and client-specific obligations vary, the firm also set guidelines covering which tasks were approved for AI use, what information could be entered, and how outputs should be reviewed. It separately reviewed contractual protections, role-based access controls, and data handling requirements. Moving to an OpenAI environment with Australian data residency reportedly let the firm expand access while still meeting internal policy and client expectations.
ChatGPT in day-to-day operations
The concrete improvements varied by department but were substantial across the board. In recruiting, research and data-extraction work dropped from about 4 hours to 20 minutes, and reference-check processing saved roughly 25 minutes per candidate. Marketing and business development use ChatGPT to organize past proposal materials and repurpose them for new opportunities—given the firm produces 400 to 500 pitches a year, this has helped with both drafting speed and consistency of tone.
Finance uses ChatGPT for spreadsheet work and data analysis, while the technology team relies on it for documentation, scripts, and training materials. Legal advice itself still runs through a separately approved platform, Harvey, with ChatGPT supporting the operational work around it. Across every use case, the responsibility for defining task scope and reviewing and approving outputs remains with humans.
| Task | Before | After |
|---|---|---|
| Recruiting research/data extraction | ~4 hours | 20 minutes |
| Conflict/KYC/AML checks | 3–8 hours | 5 minutes |
| Audit reports (300 entities) | A full day of manual work | Time saved |
Codex takes on execution
Where ChatGPT supported individual tasks, Codex went a step further, completing multi-stage operational workflows end-to-end under human review. It prepared audit reports for 300 entities, saving a full day of manual work, and handled reviewing and renaming 1,100 files for upload to another system—work that reportedly used to take several days.
For conflict checks, anti-money-laundering (AML) reviews, politically exposed person (PEP) screening, and know-your-customer (KYC) procedures, the firm built workflows where Codex handles the research and processing steps and produces reports for humans to review and approve. Some checks that used to take up to 8 hours now take 5 minutes. On the technology side, a DevOps staffer built a "watchtower" using Codex to monitor the firm's AWS environment, gathering operational signals and escalating issues to humans.
One especially notable example, mentioned by Chief Marketing Officer Daniel Quinn, is a custom GPT built as a "digital twin" of the CEO. The available sources don't provide enough detail to support this section further.
The competitive landscape for legal AI
OpenAI isn't the only company chasing law firms. As covered in Google launches Gemini for Legal to automate contracts and research, Google also previewed a product last month aimed at law firms and corporate legal departments, taking the approach of integrating with existing legal systems like iManage, DocuSign, and Harvey. Gilbert+Tobin's setup—using Harvey for legal advice and ChatGPT plus Codex for operations—fits into this same broader trend.
OpenAI says it's building out controls to expand access to ChatGPT Work and Codex, with a longer-term goal of letting employees complete tasks using approved organizational context without having to switch between systems.
Editor's take
What makes this case interesting is that the numbers come from an industry—law—that tends to be among the most cautious about adopting AI. Hitting an 87% active-use rate despite the twin constraints of confidentiality and professional liability suggests that governance design, not raw tool capability, is what really drives adoption. Set alongside OpenAI's recent finding that top-performing companies now use 8.3 times more tokens than average firms (see OpenAI reveals frontier companies' token gap widens to 8.3x), Gilbert+Tobin looks like an organization closer to that "frontier" tier.
On the other hand, OpenAI's own research found Codex adoption at just 17% among organizational subscribers and under 1% among individual subscribers (see 98% of employees use ChatGPT Work, but almost no one outside the company does)—which makes this firm's 87% figure exceptionally high by comparison. The takeaway seems to be twofold: usage wasn't mandated, and training was tailored department by department. Rather than generic training sessions, showing up in team meetings with use cases built for that specific job appears to be the key variable behind the adoption gap.
The practical lesson for domestic law firms, accounting practices, or consulting organizations is fairly clear. Instead of deploying AI directly into legal advice itself, start by automating repetitive operational work—research, data extraction, file organization—while keeping human review and approval authority intact. That sequencing can lower resistance to adoption even in departments like audit and compliance, where the cost of mistakes is high. As ChatGPT Work and Codex open up to more organizations, it should become clearer within a few months just how much execution AI can be trusted to handle in trust-dependent fields like law and accounting.
Comments