Introduction
OpenAI wants to make ChatGPT the place where in-house lawyers work, not one more tool alongside the others. That is the message of the webinar Connected Intelligence for In-House Legal, hosted by OpenAI on Thursday 29 October 2026, which we attended. For 47 minutes, the company showed an assistant that finds a contract in Ironclad, an exchange in Gmail and a line of code in GitHub, checks them against the legal department’s rules, prepares a reply and sends it. Specialist vendors, including Harvey and Legora, appear in it as plugins. [1]
This presentation is not an isolated event. Between 9 September and 5 November, Harvey and Legora raised $1.15bn, Microsoft and Anthropic launched their own legal offerings, and the law firm A&O Shearman announced a $1.1bn commitment to AI over five years. In our view, these announcements point to the same market structure: a handful of general-purpose platforms serve as the point of entry, and the value of the other players depends on what they bring to them.
Autumn 2026 in six figures
- 5,000 plugins in ChatGPT Work, across all lines of work, according to OpenAI
- +15.3 points for Astra for Law on 200 questions of US law, according to OpenAI
- $1.15bn raised by Harvey ($600m) and Legora ($550m) between 9 September and 21 October
- $1.1bn committed by A&O Shearman over five years, mostly in lawyer time
- 85% of legal departments use AI daily; 39% do not measure the return (ACC)
- 25% of executives consider their contract data ready for an agent (Pramata)
Sources: OpenAI, Harvey, Legora, A&O Shearman, ACC, Pramata. Amounts in US dollars. [1, 3, 7, 8, 9, 10, 18, 20]
1. ChatGPT becomes the point of entry to the lawyer’s software
OpenAI’s webinar starts from an observation any legal department will recognise: sales and product teams are moving faster with AI, while the information the lawyer needs remains scattered across the contract software, email, the CRM and the engineers’ tools. The proposed answer is called ChatGPT Work. According to OpenAI, this workspace already has 5,000 plugins, across all lines of work, plus those of legal partners: Ironclad, Docusign, Box, Harvey and Legora were mentioned during the session. [1]
A plugin brings two things together. Connected applications give access to a piece of software’s information and functions, within the limits of the user’s permissions. Skills are reusable instructions that describe how to carry out a task. OpenAI’s documentation specifies that a plugin may contain only skills, with no connection at all, and that installing it never bypasses the permissions set by the vendor or the company. This distinction matters for vendors: the connection gives access to the data, the skill carries the method, and the two can come from different players. [2]
To guide lawyers through this catalogue, OpenAI launched Legal Onboarding in ChatGPT Enterprise the same day. The user chooses their main practice area, for example commercial contracts. The tool then suggests the corresponding plugins (Ironclad, Salesforce, Docusign, Gmail, Google Drive, Slack) and starter tasks: reviewing the changes to a contract, getting up to speed on a matter, triaging their inbox. Activation is a matter for the workspace administrator, and each partner application keeps its own licence.
This onboarding path may seem trivial. Yet it settles the question that holds back adoption in many legal departments: where to start. It also instils a habit, that of opening ChatGPT to query one’s other software. According to a Law360 survey cited by Business Insider on 3 September, ChatGPT was already the AI tool most used by lawyers, used by 54% of respondents. OpenAI is therefore starting from existing usage and connecting the specialist tools to it. [22]
2. AI becomes useful when it reconstructs the context of a decision
Craig Bruney, a lawyer at OpenAI, handled four everyday requests without leaving ChatGPT. The table below summarises them.
Two of these demonstrations deserve a closer look, because they show what a model alone cannot do.
The first concerns a negotiation taken over while a colleague is away. The sales rep claims that the company has already agreed, for another customer, to remove the liability cap. The assistant finds this precedent and confirms that it exists. Above all, it sets out its conditions: a 90-day pilot, limited to 25 users, on synthetic or public data, approved by the VP of Legal and the chief financial officer. The contract at hand covers 900 users, real customer data and a renewable one-year term. The team’s negotiation guide then makes it possible to decide: the 45-day payment term is acceptable, the absence of a cap is not. The legal reasoning is simple. What took time was gathering the facts that make it possible.
The second concerns the UK launch of a fictional feature, Smart Routing. The lawyer does not know how it works. The assistant reads the specification document, the engineers’ exchanges in Slack and the code in GitHub, which it explains in plain language. It identifies three gaps between what is promised to customers and what the code does: data that was supposed to stay in the customer’s environment is sent to an external service, possibly outside the UK; decisions announced as reviewed by a person are automated above a certain score; the deletion of data promised within seven days takes place after thirty days. The assistant draws up an action plan from this, then drafts the note that allows UK counsel to start with the facts in hand. [1]
In our view, this second case shows where the gain lies. The lawyer no longer has to wait for an engineer to describe the product: they arrive at the discussion with a reading of the code and precise questions. The quality of their analysis then depends on two conditions that the demonstration takes for granted: that the assistant has access to the right sources, and that the company knows which one is authoritative.
This is the point stressed by Mary O’Carroll, who heads the consulting firm LECG and was speaking alongside OpenAI. According to her, what was shown works because the contracts are filed in contract management software, the negotiation guide is structured and every action leaves a trace. As she put it: “your data strategy is your AI strategy”. Contract and matter management software is therefore not disappearing. It is becoming the system of record that the assistant queries. A Pramata survey reported by Artificial Lawyer on 7 November gives a measure of the road ahead: of 464 executives surveyed, 61% say their contract data is spread across four or more systems, and only 25% consider it ready for an agent to act without human verification. [18]
3. Law firms and legal departments are building their method into tools
OpenAI’s webinar gave the floor to two organisations that have moved beyond the general-purpose assistant stage, Cooley and AT&T. What they have in common: they devote lawyers’ time to writing their rules for AI.
Richard Segal, a capital markets partner at Cooley, presented GO Public, the firm’s platform for initial public offerings. It drafts the Form S-1, the US registration statement, from the issuer’s data room and public research. Cooley has built one agent per section of the document, each with its selected precedents and its rules: more latitude for the description of the business, more rigidity for the standardised sections. In meetings, the lawyers adjust the tone or carry through a structural change, such as the addition of selling shareholders, with a simple instruction. According to Cooley’s press release of 17 September, the platform was designed with OpenAI engineers and produces a first draft “in minutes rather than days”. [1, 6]
Richard Segal drew a conclusion from this that applies beyond capital markets. Getting an AI to produce a document that looks like an S-1 is easy. Getting a high-quality S-1 is hard, and the difference comes from the firm’s judgement, built into the rules and precedents it has chosen. He adds that the result must be better, and not merely faster, failing which the firm is not making full use of the technology.
A&O Shearman’s announcement of 5 November follows the same logic, on a different scale. The firm, born of the combination of Allen & Overy and Shearman & Sterling, is committing to devote $1.1bn to AI over five years. According to its press release and the Financial Times, most of this amount corresponds to lawyer time: each practice will devote at least 3% of its billable time to building specialised tools in its field. The share of technology spending has not been disclosed, as the website Nonbillable pointed out. It is therefore an internal effort given a monetary value, not a budget for buying software. Set against the firm’s $3.7bn in revenue, it represents around $220m a year, or nearly 6% of revenue. [10, 11, 12]
David Wakeling, who heads AI at A&O Shearman, told the Financial Times that the challenge is not to keep up with every new model, but to build the teams’ expertise into specific AI systems. The firm says that 96% of its partners were using its tools in September and that usage of its ContractMatrix range, sold to its clients, has grown ninefold in a year. The Financial Times noted as early as 3 September that several international firms are building their own tools to differentiate themselves, as Freshfields is doing with Anthropic and Google. [25]
On the corporate side, AT&T is following a comparable path with LegalEdge, its AI-based legal centre of expertise, launched in January 2026. In the video shown by OpenAI, the request reaches the lawyer in charge with the history of the matter and suggested first steps, and the decision remains with the lawyer. According to AT&T’s press release of 8 October, the company has become OpenAI’s design partner for legal departments. [5]
These three examples presuppose a skill that Mary O’Carroll named: the legal engineer, able to describe a legal task, break it down and define what can be automated. She recommends assigning a person or a partner to it, and planning resources for the long term, because a procedure written for AI has to be maintained when the rules, the tools or the organisation change. We observe that this skill is lacking in most legal teams. It is an opening for legaltechs that know how to support their clients, and not merely deliver software to them.
4. Astra for Law is a reminder that research is only as good as its corpus
OpenAI used the webinar to open Astra for Law to in-house legal teams. The tool had been launched on 17 September for a first group of law firms. It is not a new model. It is a configuration of GPT-6 Astra to which OpenAI adds an index of more than 230 million addresses of US legal sources, instructions for analysis and drafting, and a setting that gives the model more reasoning time. According to OpenAI, the built-in case-law collection covers more than 99% of published US decisions with precedential value. [1, 3, 4]
The result highlighted by OpenAI is as follows: on 200 legal research questions from Vals AI’s private validation set, Astra for Law passes 54.0%, against 38.7% for GPT-6 Astra equipped with web search alone.
Source: OpenAI, 17 September 2026. Share of the 200 questions in Vals AI’s private validation set passed on the overall accuracy criterion, maximum reasoning effort. Chart by Leaid. [3, 4]
Three readings stand out. The 15.3-point gain comes from the index and the instructions, not from the model, which is the same in both cases: the quality of AI legal research depends first of all on the corpus being queried. Next, a rate of 54% means that nearly one question in two is not passed under the test’s criteria: the tool speeds up research and does not remove the need to verify it. Finally, the index and the evaluation concern US law. OpenAI has announced nothing for French or European law.
The example shown during the webinar illustrates the intended use. The task was to find the Delaware precedent closest to an asset sale approved by a board of directors over a holiday weekend, after receiving hundreds of pages, under the pressure of a deadline that the buyer had itself requested. The answer identifies the decision OTK Associates LLC v. Friedman and flags what distinguishes it from the case submitted: there, the independence of the directors was at issue. OpenAI specified that the answer had been prepared in advance, because such a search takes around ten minutes and several dozen to several hundred queries.
In France, September’s deals follow the same logic. Septeo acquired Lexbase on 3 September, RELX completed the acquisition of Doctrine on 4 September, and Le Monde du Droit announced on 8 September the takeover of Query Juriste by Lexroom. Players that hold a French or European corpus have the asset that the model providers have not built for these bodies of law. [23, 24]
5. Autumn 2026 sets three general-purpose platforms against the specialists
OpenAI is not alone in addressing lawyers directly. Microsoft announced Copilot for legal on 15 October: more than twenty legal skills in Word and Copilot, and 25 partner plugins, including iManage, NetDocuments, Docusign, Harvey and Legora. On 28 October, Anthropic presented a gallery of ready-to-adapt legal tools, for checking citations, comparing contracts or tracking matters, and claims more than 60 legal connectors since May. The three offerings follow the same pattern: a general-purpose interface, preconfigured legal methods, and specialist vendors distributed as plugins. [13, 14]
The specialists are not standing still. Harvey raised $550m on 9 September, at a valuation of $15.5bn, then increased this round to $600m on 24 September. Legora raised $550m on 21 October, at a valuation of $8.8bn. The Swedish vendor reports more than $200m in annual recurring revenue and says that legal departments account for more than 40% of its new customers in the third quarter. According to Artificial Lawyer, the two companies have reportedly begun preparatory discussions with banks with a view to an initial public offering, which Legora has not commented on. [7, 8, 9, 17]
Sources: Harvey, Legora, A&O Shearman, Financial Times, Nonbillable. Amounts in US dollars, not to be added together. Table by Leaid. [7, 8, 9, 10, 11, 12]
These two vendors occupy a particular position: they are present in the catalogues of OpenAI, Microsoft and Anthropic, and at the same time they are building their own environment. On 13 October, Legora launched Intake, which receives requests addressed to the legal department in Slack, then on 2 November announced an integration with Conga in which Conga remains the system of record for contracts. In early November, Harvey integrated Jus AI, the Jus Mundi tool specialising in international arbitration. The specialists are thus connecting to systems of record and to corpora, exactly as the general-purpose platforms do. [15, 16, 18]
Buyers, for their part, are asking for accountability. The Financial Times reported on 1 September that Goldman Sachs, Morgan Stanley and Citigroup are asking their law firms for fees to reflect the time saved thanks to AI. The Association of Corporate Counsel survey published on 29 September, conducted among 576 in-house counsel, indicates that 85% of legal departments use AI daily, that 39% do not measure its return on investment, and that the purchasing decision lies with another department, most often IT, in 73% of organisations. According to Thomson Reuters’ Future of Professionals 2026 report, 71% of in-house counsel expect their law firms to change their business model with AI. [19, 20, 21]
In our view, three lessons stand out:
- The working interface is shifting to the general-purpose platforms. ChatGPT, Copilot and Claude each offer a space from which the lawyer drives their software. A product that is not accessible there will have to be opened separately, which the demonstrations present as a waste of time.
- Value is concentrating in three assets. Systems of record and corpora, the working method, and the ability to execute an action reliably and traceably. OpenAI’s demonstrations rest entirely on these assets, which OpenAI does not hold.
- Method is becoming an investment. Cooley, A&O Shearman and AT&T devote lawyers’ time to writing their rules for AI. The demand for support concerns this work as much as the software.
Conclusion
We are convinced that OpenAI’s offering opens up more space for legaltechs than it closes off, provided they choose their place. A founder must decide what their product exposes in ChatGPT, Copilot and Claude, for what exactly it wants to be the authoritative source, and how it sells its method along with its software. For French and European players, local law remains an asset: the Astra for Law index is American, and a published evaluation on questions of French law would have few equivalents today.
For legal departments and law firms, the work begins before the tool. It consists of knowing which source is authoritative, describing a first process with its steps and validation points, and setting limits on what the assistant can send or modify. For lawyers in private practice, this framework comes on top of professional secrecy, which determines which data can be connected. Organisations that have formalised their method will get the most out of any platform; the others will find their scattered data waiting for them there.
Sources
- OpenAI, webinar “Connected Intelligence for In-House Legal”, 29 October 2026: transcript read in full and replay (47 minutes) webinar.openai.com/connected-intelligence-for-in-house-legal/
- OpenAI Help Center, “Plugins in ChatGPT”, accessed 10 November 2026 help.openai.com/en/articles/20001256-plugins-in-chatgpt
- OpenAI, “Introducing Astra for Law”, 17 September 2026 openai.com/index/astra-for-law/
- SiliconANGLE, “OpenAI launches Astra for Law, a GPT-6 configuration for legal research”, 17 September 2026 siliconangle.com/2026/09/17/openai-launches-astra-for-law-a-gpt-6-configuration-for-legal-research/
- AT&T, press release on the collaboration with OpenAI and LegalEdge, 8 October 2026 about.att.com/story/2026/att-legaledge.html
- Cooley, “Cooley Launches GO Public With OpenAI”, 17 September 2026 cooley.com/news/coverage/2026/2026-09-16-cooley-launches-go-public-with-openai
- Harvey, announcement of the $550m raise at a $15.5bn valuation, 9 September 2026 harvey.ai/blog/harvey-raises-dollar550m-at-a-dollar155b-valuation-to-help-legal-teams-own-their-intelligence
- Harvey, investment by Ontario Teachers’ and extension of the round to $600m, 24 September 2026 harvey.ai/blog/harvey-announces-investment-from-ontario-teachers-pension-plan-and-deepens-its-commitment-to-canada
- Legora, announcement of the $550m Series E at an $8.8bn valuation, 21 October 2026 legora.com/newsroom/legora-raises-550-million-series-e-at-8.8-billion-valuation-to-accelerate-growth
- A&O Shearman, press release on the $1.1bn commitment, 5 November 2026 aoshearman.com/en/news/ao-shearman-accelerates-ai-strategy-with-a-usd11bn-lawyer-led-commitment
- Financial Times, “A&O Shearman invests $1.1bn in AI”, 5 November 2026 ft.com/content/b832654a-79d1-4db0-9de8-d1a65251539d
- Nonbillable, “A&O Shearman ramps up AI push with five-year plan”, 5 November 2026 nonbillable.co.uk/news/ao-shearman-ramps-up-ai-push-five-year-plan
- Microsoft, “Introducing Microsoft Copilot for legal”, 15 October 2026 techcommunity.microsoft.com/blog/microsoft-copilot-blog/introducing-microsoft-copilot-for-legal/4564246
- Anthropic, “Introducing artifacts for legal work”, 28 October 2026 claude.com/resources/articles/introducing-artifacts-for-legal-work
- Legora, “Legora launches Intake”, 13 October 2026 legora.com/newsroom/legora-launches-intake-an-agentic-front-door-for-in-house-legal-teams
- Legora and Conga, announcement of their integration, 2 November 2026 legora.com/newsroom/legora-and-conga-integrate-to-bring-enterprise-contracts-into-the-agentic-operating-system-for-legal-work
- Artificial Lawyer, “Harvey + Legora IPO Plans, A&O $1.1bn, Legal Innovators NY”, 6 November 2026
- Artificial Lawyer, “Bonus Wrap: Harvey, Jus Mundi, LI NY +”, 7 November 2026 (Harvey and Jus Mundi partnership, Pramata survey)
- Financial Times, article of 1 September 2026 on the requests made by Goldman Sachs, Morgan Stanley and Citigroup to their law firms ft.com/content/5240a6ac-b2e8-4897-a0a4-cbc7fc283bc9
- Association of Corporate Counsel, “The In-house Legal AI Landscape”, 29 September 2026 acc.com/about/newsroom/news/association-corporate-counsel-survey-finds-85-house-legal-departments-have
- Thomson Reuters Institute, “Future of Professionals Report 2026” thomsonreuters.com/en/institute/future-of-professionals-2026/report-legal
- Business Insider, “OpenAI is about to make its biggest push yet into the legal industry”, 3 September 2026
- Septeo, press release on the acquisition of Lexbase, 3 September 2026
- Le Monde du Droit, “Le Groupe RELX finalise l’acquisition de Doctrine”, 4 September 2026, and “IA juridique : Lexroom rachète la jeune pousse française Query Juriste”, 8 September 2026
- Financial Times, “Law firms seek bespoke differences in legal AI”, 3 September 2026