Hat tip to a risk expert colleague who noted an interesting “frontier” spotlight of conflicts, client due diligence, and compliance in this AI case study: “How law firm Gilbert + Tobin governs and scales AI with OpenAI” —

• “Gilbert + Tobin used [OpenAI] Codex to help build an AI-enabled workflow for selected conflict, anti-money laundering, politically exposed person, and know-your-customer checks. It completes research and processing steps, then produces a report for human review and sign-off. Selected checks that previously took up to eight hours can now be completed in minutes.”
• “When generative AI began reshaping knowledge work, Gilbert + Tobin saw an opportunity—not to change how it delivers legal advice, but to raise the standard of the operational work that supports it. The firm set out to give every operations team sharper tools for research, drafting, and analysis, while preserving the governance and professional accountability its clients expect.”
• “ChatGPT Enterprise was introduced first to operations teams, where practical use cases could be tested in a controlled environment. From there, adoption expanded into marketing, business development, recruitment, finance, technology, business transformation, and parts of the legal practice. Throughout, the focus has remained on how the firm operates, not on displacing the professional judgment at the heart of its legal services.”
• “Gilbert + Tobin paired that leadership support with role-specific enablement. Instead of relying on generic training, the business transformation team joined individual team meetings to demonstrate workflows tailored to marketing, finance, recruitment, operations, and other functions. Internal videos, showcases, challenges, and use-case stories helped employees keep pace as ChatGPT gained new capabilities.”
• “Gilbert + Tobin handles confidential client and business information, including information subject to different client requirements. Its AI rollout was supported by clear guidance on approved tasks, what employees could enter, and how outputs should be reviewed. The firm also assessed contractual protections, role-based access, data-processing requirements, and administrative controls.”

“Latham & Watkins AI Adoption: How BigLaw Is Deploying Generative AI” —

• “The firm has stood up a Generative AI Task Force, deployed Harvey across thousands of attorneys, rolled out Microsoft 365 Copilot, published external client guidance on AI risk, and built mandatory training into associate development.”
• “The use case BigLaw firms talk about least and undervalue most is conversational client intake and matter onboarding. Latham, like every AmLaw firm, runs a meaningful intake process every time a new matter opens — conflicts checks, engagement letter drafting, scope definition, billing arrangement capture, KYC where relevant, and matter-team selection. Today, most of this happens through a combination of email threads, Word documents, intake portals, and a back-and-forth between business development, the originating partner, the conflicts team, and the new business intake group.”
• “That entire workflow is a conversation pretending to be a form. Engagement letter intake portals demand structured fields (‘matter type,’ ‘billing arrangement,’ ‘estimated fees,’ ‘key parties’). But the signal a client gives over email is unstructured — ‘this is a complicated cross-border thing, we expect to see a counter-bid in two weeks, I need someone who’s worked with the German regulator before, the budget is whatever it has to be.’ Forms flatten that signal. Senior partners then spend their own time translating client narrative into the structured intake fields.”
• “Conversational AI is the right instrument here. An AI intake agent can ask the natural follow-ups a senior associate would — ‘you mentioned a counter-bid; do you want us to start preparing defensive materials now or wait?’ — and capture the matter context in the client’s own language before any structured fields are populated. The structured fields then get derived from the conversation, not demanded up front.”

“Law firm intake that runs the conflict check before anyone calls the prospect back” —

• “An intake workflow for law firms that turns web forms, chats, emails and voicemails into structured intake: matter type, parties, jurisdiction and urgent deadlines. It searches the practice management system for conflicts across the parties and their corporate groups, triages each inquiry against the practice groups’ criteria, and prepares the engagement letter and a consultation slot. A person clears every possible conflict and a lawyer accepts every matter. The assistant never gives legal advice and says it is an AI.”

And in terms of OCGs, one consultant hypothesizes: “Ken Crutchfield: Taking BigLaw Private? The Implications Of Kirkland’s Decision To Stop Disclosing Financials” —

• “Kirkland & Ellis made headlines again when it stated it will no longer disclose financial information to The American Lawyer magazine and other media after 2026. Firm leadership said financial disclosures could put too much focus on quantitative metrics over the quality and caliber of the firm’s legal services.”
• “Clients will still care about financials and reputation, but their interests aren’t necessarily predicated on profit per equity partner and other top-level metrics reported publicly in an Am Law list. They require compliance with Outside Counsel Guidelines (OCGs) to manage supplier risk and pricing, and compliance.”
• “The scope of those guidelines continues to expand, and they now can cover billing, staffing, technology, security, and AI use. It is not difficult to imagine negotiated disclosures of financial metrics privately as a necessary condition of the ongoing relationship.”

“The AI Agent Privacy Problem: Who’s Liable When Your Legal AI Shares Client Data?” —

• “AI agents are no longer a future-facing concept in legal practice. They are already drafting motions, reviewing contracts, conducting due diligence, managing calendars and email, and in some deployments taking multi-step actions across systems. They process some of the most sensitive data that exists: client communications, litigation strategy, financial records, and personal identifying information. Adoption continues to outpace the guardrails designed to protect the data those agents touch.”
• “That gap is still where the liability lives, and the last several months have sharpened the contours of it.”
• “In June 2026, the American Arbitration Association and Integra Ledger announced the Legal Context Protocol, an open standard intended to attach discoverable, verifiable legal terms, consent, and dispute-resolution pathways to agent-to-agent commercial transactions. Founding contributors included Google, IBM, Circle, Wayfair, and others. The protocol remains a useful piece of infrastructure for agentic commerce. It does not, however, resolve the distinct confidentiality, privilege, and professional-responsibility problems that arise when agents operate inside legal workflows.”
• “When a lawyer uses a traditional software tool, the workflow is relatively clear: the attorney inputs information, the tool processes it, and the attorney reviews the output. Privilege and confidentiality obligations run through the attorney’s conduct. AI agents disrupt that model. An agent tasked with drafting, reviewing, or conducting diligence does not simply process information. It can ingest it, retain it (at least temporarily), route it through third-party infrastructure, and in some configurations act on it sending messages, updating records, or accessing systems before meaningful human review occurs.”
• “ABA Model Rule 1.6 still requires attorneys to make reasonable efforts to prevent the inadvertent or unauthorized disclosure of information relating to a representation. ABA Formal Opinion 512 (2024) remains the foundational ethics guidance: lawyers must understand the specific tool’s capabilities and limitations, evaluate disclosure risks before inputting client information (especially into self-learning systems), and obtain informed client consent where the tool’s design creates a material risk of disclosure. Boilerplate engagement-letter language is not sufficient. State bars have continued to apply and refine this framework. Oregon’s Formal Opinion 2026-208, for example, addresses chatbots and AI agents used for marketing and intake, emphasizing competence, supervision, and the lawyer’s ongoing responsibility for accuracy and confidentiality. Florida and other jurisdictions have issued parallel guidance.”
• “The most concrete statutory development arrived at the end of September. On Sept. 30, 2026, California Governor Gavin Newsom signed SB 574, the first state statute specifically regulating generative AI use by attorneys, arbitrators, judicial officers, and alternative dispute resolution providers. Effective Jan. 1, 2027, the law:”
• “When an AI agent generates a contract summary, flags litigation risk, prioritizes documents, or drafts a demand letter, the question whether that output constitutes or materially influences an automated decision with a significant effect on a data subject is no longer purely theoretical. In some jurisdictions it may trigger notice, opt-out, access, or human-review rights that most current legal AI deployments are not designed to honor. Law firms and legal departments that treat these obligations as ‘just another vendor issue’ are exposing themselves and their clients.”
• “Vendor agreements data processing addenda, confidentiality clauses, and terms-of-service representations remain necessary but still not sufficient. They govern the firm-provider relationship. They do not create rights for clients whose data is processed, nor do they substitute for the attorney’s independent ethical and legal duties. Attorneys deploying agents must still ask, and document, the practical questions: Where does the data go? Who can access it? Is it used for training? What is the retention period? What happens in a breach? What permissions does the agent have, and can those permissions be tightly scoped and monitored?”
• “The last few months have also shown that agent behavior can exceed intended scope. Documented incidents involving OpenAI and other developers’ agents attempting or achieving unauthorized access to external systems prompted a Sept. 30 Senate hearing and, on Oct. 1, the announcement of the bipartisan AI Agent Accountability Act by Senators Josh Hawley and Chris Murphy. We’ve seen a rise in law firm data breaches meanwhile News Nation and Fox Business News had me on air to discuss the ‘pacing of the frontier’ and the AI risks ahead. These new proposals would extend Computer Fraud and Abuse Act liability to both operators (for knowingly operating an agent that recklessly causes hacking damage or loss) and developers (for failing to implement reasonable safeguards when they knew or had reason to know of the agent’s hacking capabilities). It would also empower the attorney general and state attorneys general to seek injunctions. Parallel proposals such as the CLAIM Act discussion draft seek to clarify civil liability when agent conduct would be tortious or criminal if performed by a person. These bills are not yet law, but they signal that ‘the AI did it’ is unlikely to be an accepted defense, and that both developers and deployers face increasing scrutiny.”
• “Privilege risks have also moved from theoretical to concrete. Courts have held that materials generated through consumer-grade tools whose terms permit retention or third-party access can destroy attorney-client privilege and work-product protection. The reasonable expectation of confidentiality simply does not exist under those terms.”
• “Technical protocols such as the Legal Context Protocol address one slice of the agentic economy commercial transactions. Professional responsibility rules, state statutes like California’s SB 574, evolving privacy and ADMT requirements, and emerging liability proposals address another. Courts have not yet issued definitive rulings on the full range of agent-related confidentiality and privilege questions, but the direction of travel is clear: the lawyer or firm that deploys the agent remains responsible.”
• “Attorneys who build documented data-governance frameworks around their AI tools—restricted-access environments, least-privilege permissions, human review gates for consequential actions, client disclosure where required, and verification protocols will be better positioned than those who treat agents as ordinary software and wait for a court or bar to define the boundaries after an incident. The liability question is no longer purely open. The answers are beginning to arrive, and they continue to point to the human professionals who choose to deploy the tools.”