ChatGPT for legal work, explained

ChatGPT is OpenAI's general-purpose assistant. It writes, summarises, translates and reasons over whatever you give it, for anyone who asks. Lawyers use it the way everyone else does: in a chat window, one conversation at a time.

For organisations, OpenAI sells ChatGPT Business (formerly Team) and ChatGPT Enterprise. Business is reported at about $20 per user per month on annual terms, or $25 monthly, with a minimum of two seats. Enterprise has no public price. Both are reported to exclude your data from model training by default.

OpenAI has no legal-specific product. In June 2026 it hired Jason Boehmig, co-founder of the contract platform Ironclad, to lead work on products for the legal sector. Nothing has shipped. ChatGPT for law, today, means the same assistant everyone else uses.

• First drafts. A starting point for an email to the business, a policy outline or a cover note, rewritten by the lawyer before it goes anywhere.
• Summaries. Turning a long public document, such as a consultation paper or a regulator's guidance, into a page of notes.
• Plain English. Explaining a clause to a non-lawyer in the business, or rewriting legalese for a staff handbook.
• Thinking out loud. Listing the issues a type of agreement usually raises, or the questions to ask before a negotiation.
• Tidying. Fixing tone, shortening a memo, or turning bullet points into a paragraph.

Paste in a contract and ask ChatGPT what is unusual, and it will give you an answer. For a quick read of an unfamiliar agreement, that has its uses.

A contract review in an in-house team is more than an answer. It is a check against the team's positions: our cap on liability, our governing law, our fallback when the other side won't move. It is a redline the counterparty can read, a note of what was conceded, and a record of the signed terms for the day the renewal comes round.

ChatGPT knows none of your positions unless someone types them in, in that conversation. It doesn't keep the team's playbook, doesn't queue the contracts waiting for review, and doesn't file what was agreed. Every lawyer who uses it starts again, with their own prompt, and their own idea of the standard.

So, can ChatGPT review contracts? It can help one lawyer read one. The review, the redline and the record are still left to the lawyer.

In Mata v. Avianca (S.D.N.Y. 2023), lawyers filed a brief citing cases that ChatGPT had made up. The court sanctioned them $5,000. The case became the standard warning about generative AI in legal work.

In England, the Divisional Court dealt with the same problem in Ayinde v London Borough of Haringey and Al-Haroun v Qatar National Bank [2025] EWHC 1383 (Admin), decided together on 6 June 2025. Both involved fictitious citations put before the court. The judgment is a reminder that responsibility for checking every authority sits with the lawyer, whatever tool produced it.

The cause is the same in each case. A general model is built to produce fluent, likely-sounding text. Without a connection to a verified source, a case name that sounds right is, to the model, as good as one that exists.

ABA Formal Opinion 512 (29 July 2024) sets out how US lawyers' existing duties apply to generative AI. Lawyers must verify output in proportion to the task. They need informed client consent before putting client information into self-learning tools. Firms need policies and training, and must vet vendors for security and confidentiality.

The SRA's warning notice on the misuse of AI (17 August 2026) is sharper. Client data should go into AI tools only where "contractual, technical and organisational safeguards" exist. Privilege "may be permanently waived" through public tools. Firms need "effective governance structures, systems and controls". The SRA received 42 reports of AI misuse between July 2025 and July 2026.

Neither bans ChatGPT. Both make the same point: a general assistant used ad hoc, by individuals, without controls, is the risk. The safeguards are the organisation's job, not the tool's.

1. Use a business plan, not a personal account. ChatGPT Business and Enterprise are reported to exclude your data from training by default. Personal accounts are the wrong place for any work material.
2. Keep privileged and client material out. Unless your contract with OpenAI and your own controls meet the safeguards the SRA describes, don't paste in privileged advice, client data or live deal documents.
3. Never rely on a citation. Check every case, statute and quotation against a primary source before it goes anywhere. Treat any authority ChatGPT gives you as a lead, not a reference.
4. Keep it to drafts. Use it for first drafts, summaries of public material and plain-English rewrites. A lawyer reads and owns everything that leaves the team.
5. Write it down. Set a short team policy: which plan, which tasks, what never goes in, and who checks. ABA Opinion 512 and the SRA both expect one.

In-house legal work arrives from the business, all day, in Slack and email. A customer's paper from sales. A supplier MSA from procurement. Twenty NDAs a week. Each one needs picking up, checking against the team's positions, returning as a redline, filing, and remembering at renewal. A chat window forgets the conversation the moment you close it.

ChatGPT waits for someone to open it and ask. What it gives back is an answer for that person to turn into work. The playbook lives in their head. The record lives nowhere.

Claren is built the other way round. Its envoys are AI employees, each with its own computer and its own memory. They take requests from the Claren app or straight from Slack. They review against the team's single playbook and hand back tracked-change redlines in Word, with a comment on every change, your fallback positions applied and a summary of what is off-playbook. They file every contract and its terms in the team's datastore. They run routines on a schedule, or when a message lands. Before anything leaves the team, they ask.

And the controls regulators ask for come built in. Claren holds SOC 2 Type II and ISO/IEC 27001:2022, keeps each organisation's data apart, keeps an audit trail of every user action and hosts in the US, UK or EU. Neither Claren nor its model providers train on your data.

1. Is the task low-stakes drafting, or legal work? For a first draft of an email or a summary of a public paper, ChatGPT is fine. For contracts the business will sign, you want a reviewer that follows your playbook every time. That is Claren.
2. Where do requests come from? If they come from the business, in Slack and email, you want something that picks them up and tracks them there. Claren's envoys do; ChatGPT waits to be asked.
3. Whose standard applies? In ChatGPT, whatever each user types in. In Claren, the team's playbook, written once and followed by every envoy.
4. Could you show a regulator your controls? The SRA asks for governance, systems and controls. Claren gives you sign-off before anything leaves the team, an audit trail, data kept apart and SOC 2 Type II and ISO 27001 behind it.

There's no need to give up OpenAI's models. Claren's envoys run on models from OpenAI, Anthropic and Google, chosen per envoy, through their APIs, and Claren holds zero data retention agreements with OpenAI and Anthropic. A team that likes how ChatGPT writes can have an envoy that writes the same way, inside a legal workflow. Claren for ChatGPT, which brings your envoys and records into ChatGPT itself, is coming soon.

A common split: lawyers keep ChatGPT for quick questions and personal drafting, and route the work that repeats (NDAs, supplier contracts, renewals, data room reviews) to Claren's envoys, which pick it up from Slack and hand it back done.