The Spotlight this issue is a debate rather than a single development. In one week, the heads of the leading AI labs called for a deliberate slowdown in frontier development, the White House rejected the idea of new federal rules, and the largest chip and platform companies argued that no regulation is needed at all. That leaves a business lawyer with a practical question, and we take it up at the top.

Below that, the usual format: recent court decisions, new laws and regulations, ethics guidance, and practical resources, grouped by topic so you can go straight to what matters for your practice. Every item links to a source where you can read more.

The Task Force welcomes your participation. If you come across an article worth sharing, have an idea for the Task Force, or want to get involved, we want to hear from you. You can find us at https://flabizlaw.org/committees-task-forces/task-forces/artificial-intelligence-task-force/

— The BLS AI Task Force

The frontier labs asked for a slowdown; the government and the chipmakers said no

Dario Amodei (Anthropic); with reporting from Forbes, Axios, CNBC, TechCrunch, and Insurance Business

Why it matters: in a single week the question of whether frontier AI should be reined in went from an in-house safety debate to a live political and commercial fight, and the near-term answer at the federal level is no new rules. That reframes where the work is for anyone advising on AI. Not federal legislation, but state law, private contracts, and ordinary liability.

On September 12, Anthropic CEO Dario Amodei published “We Must Pace the Frontier”, arguing that the leading labs should deliberately slow the rate at which model capabilities advance — not pause development — so that safety and oversight can keep up, and committing Anthropic to give independent evaluators like METR permanent, employee-level access to check its safety practices (Forbes). Within hours OpenAI’s Sam Altman said he agreed, and xAI’s Elon Musk added “Dario is right,” a rare alignment among three rivals. The call also drew support from outside the labs: Geoffrey Hinton, the computer scientist often called the “godfather of AI,” backed Amodei’s warning and rejected the argument that such warnings are a marketing play, noting a pause would cut the value of the companies issuing them (Insurance Business).

The response went the other way, fast. President Trump rejected new AI rules, framing the race as one the country cannot afford to lose — “whoever wins AI wins” — and White House AI czar David Sacks said the labs should police themselves rather than ask the government to do it (Axios; CNBC). Then the infrastructure and platform side weighed in: Nvidia’s Jensen Huang said safety is “an engineering problem, not a legal one” and that the industry does not need “any new laws” or “new regulations,” and Meta’s Mark Zuckerberg sided with him, arguing that market forces and the liability labs already face push them to prevent harm.

For members, the practical read is about the regulatory baseline, not the merits. The most capable AI companies are split on whether frontier development needs outside limits, and the federal executive branch has signaled it will not impose them soon. That leaves the enforceable obligations where clients actually meet them — state statutes like California’s new auditor framework and the AI-agent bills below, contractual allocation of AI risk, and ordinary tort and product liability — not a coming federal frontier-AI regime. For now, the administration has signaled that new federal frontier-AI rules are not a near-term priority.

A D.C. appeals court struck a bank’s brief over four AI-hallucinated cases, then found it had no power to do much more

District of Columbia Court of Appeals (Douglas v. Deutsche Bank National Trust Co., No. 24-CV-1099, Sept. 3, 2026); reported by the ABA Journal

The court struck a mortgage servicer’s entire appellate brief after four of its cited authorities turned out not to exist, the product of an attorney’s unverified use of Google’s generative AI search tool. The panel said every attorney who signed the brief bore some responsibility, not just the one who ran the search, and referred the matter to the Office of Disciplinary Counsel. The part worth reading is the sanctions discussion: the court concluded that its own appellate rules did not clearly authorize it to fine or otherwise sanction counsel beyond striking the brief, and referred that open question to its Rules Committee, a reminder that appellate sanctions authority for AI hallucinations is still unsettled and uneven from court to court.

California’s legislature moves to make lawyer AI use a matter of statute, not just a bar rule

Sullivan & Cromwell LLP

SB 574 has passed both houses and sits on Governor Newsom’s desk; if signed, California becomes the first state with a statute, rather than a bar rule, governing lawyers’ use of generative AI. It makes it an attorney’s duty not to enter confidential or nonpublic information into a public generative AI system, or into any AI tool that does not restrict access to the information to the lawyer and other authorized, confidentiality-bound users, and it amends the state’s verification rules so the attorney responsible for a filing must personally confirm every citation, AI-generated or not, with no room to push that check onto a junior associate or paralegal. Florida lawyers with California matters or co-counsel should track whether it is signed, and it previews the statutory model other states may copy.

California creates a first-in-the-nation registration framework for AI auditors

Kelley Drye & Warren LLP (Ad Law Access)

Governor Newsom signed SB 813 and AB 1405 on September 9, standing up the country’s first state process to recognize and register independent AI “verification organizations,” with a public registry and, from 2029, a bar on unregistered persons conducting a covered audit. This is not an audit mandate. Illinois’s AI Safety Measures Act already requires third-party audits of frontier models. California’s first is the credentialing-and-registration layer around who may perform them. Members whose clients build or buy higher-risk AI should watch whether a state-recognized auditor credential becomes a practical expectation, as SOC 2 did in security.

A bipartisan House bill would make companies keep a running inventory of their AI agents

Axios

Representatives Josh Gottheimer and Mike Lawler introduced the Stop Rogue AI Act, which would require any company running AI agents to keep a continuous, machine-readable inventory of them and give NIST a year to write deployment standards covering verification, security testing, tamper-proof action logs, and vendor identification. Compliance would be voluntary in general but mandatory for federal contractors, one of several congressional proposals addressing AI agents this year. Even if it never passes, the bill sketches what a defensible AI-agent deployment is expected to look like. A useful reference for advising any client putting agents into production.

3M’s expert leaned on ChatGPT to argue zero fault, and the prompts became evidence

Above the Law (Joe Patrice)

A Harris County, Texas jury awarded $61.5 million against 3M in one of the trials over the 2020 Watson Grinding explosion, finding the company partly at fault despite its expert’s use of ChatGPT. The defense engineer had prompted the tool to help “show how 3M is 0% at fault,” and the plaintiffs obtained roughly 350 pages of those prompts in discovery, exposing AI’s fingerprints across most of his report. The verdict did not turn on the AI use, but the case clearly illustrates that an expert’s prompt history can become discoverable when it feeds the expert’s opinion and report, not just the polished report a client eventually sees

A federal judge bans AI-generated text outright and makes every filing certify a human wrote it

U.S. District Court for the Northern District of Texas — Senior Judge Jane J. Boyle, judge-specific requirements § I.K

Effective September 1, Judge Boyle’s standing requirements bar any AI-generated text from any pleading, motion, brief, or other filing an attorney makes before her: every word must be drafted and checked for accuracy by a human attorney, and every filing must carry a signed certificate that the text was generated in its entirety by a human, or the filing may be stricken. Using AI only to identify errors is allowed but must be disclosed under Local Rule 7.2(f), with any resulting edit made and verified by a human; pro se parties are held to the same standard, and noncompliance can draw litigation or Rule 11 sanctions. It sits at the strict end of a widening spectrum: plenty of judges now require AI disclosure or verification, but a flat ban plus a human-authorship certificate on every filing is a materially higher bar, and any lawyer with a matter in that court needs to know it before the next filing.

Shareholders are suing boards over how they adopted AI, not just over the AI itself

Ropes & Gray LLP (William Michener, David M. McIntosh, Joshua Talicska, and Toby Shao)

A wave of shareholder derivative complaints filed this spring and summer — against the directors and officers of Adobe (N.D. Cal.), Microsoft (W.D. Wash.), and Nvidia (N.D. Ill.) — allege that leadership breached fiduciary duties by approving AI strategies built on training data the plaintiffs say was pirated, and by making governance disclosures the plaintiffs say the companies’ actual practices contradicted. A notable feature of the derivative posture is that the claims do not require proving the underlying infringement; several are pled from the gap between a public commitment (Adobe’s “we do not mine content from the web to train Firefly,” for one) and what the company allegedly did. The takeaway for anyone advising a board: public AI commitments — proxy statements, codes of conduct, even a product-page line — can become the benchmark a plaintiff pleads the board against, and thin or undocumented board-level AI oversight is itself among the pleaded allegations.

The FDA starts mapping how it will regulate generative-AI medical devices

Womble Bond Dickinson (US) LLP (Jennifer German)

In August, the FDA’s Digital Health Center of Excellence released a discussion paper on regulating generative-AI-enabled medical devices and opened a comment docket (FDA-2026-N-7874). It imposes no new requirements, but it names the problem a device that can give different answers to the same question — and that can change after launch when a third-party model is updated — poses for a testing regime built for fixed software: how much testing is enough, when a post-market model change should trigger reassessment, and how a manufacturer oversees the outside AI its product depends on. The paper is expressly for discussion and proposes no policy change; comments are due October 19. For members with life-sciences, medical-device, or digital-health clients, the practical move is to start building performance evidence and a vendor-oversight record now, including contract terms that guarantee notice and access when the underlying model changes—while the FDA is still developing its approach.

The ILTA survey puts hard numbers on where firms your size actually are on AI

International Legal Technology Association (ILTA); reported by eDiscovery Today (Doug Austin)

ILTA’s 2026 technology survey, drawn from 508 firms, found 94% now using or exploring generative AI, up from 80% a year ago, with adoption running from 85% at the smallest firms (under 50 lawyers) to 100% at the largest. Formal governance — written policies, human review of AI output, approved-tools-only rules, required training, and citation-verification mandates — lags well behind adoption and thins out at smaller firms, and generative AI appeared for the first time on the survey’s tracked security threats, landing second behind user behavior. The firm-size breakouts give a member a concrete answer to a client’s “is our AI posture normal for a firm our size” rather than generic advice.

Morgan & Morgan commits $1 billion to AI and plans to sell its own platform to other firms

Reuters (Sara Merken); reported by the ABA Journal

Morgan & Morgan, the largest U.S. personal injury firm, said it will commit at least $1 billion to AI and technology over the next decade and will start selling its in-house platform, built at a reported $300 million, to outside firms in 2027. Founder John Morgan’s framing is the notable part: he argued that trial-focused firms are more insulated from AI than firms billing by the hour for document review and drafting, “the practices AI will replace almost entirely.” It is a data point on where large-firm AI spend and the billable-hour question are heading, and the piece notes that Morgan & Morgan attorneys were sanctioned last year for AI-hallucinated citations, though the court declined to sanction the firm itself, a reminder that the spend does not remove the verification duty.

Latham is the first big firm publicly known to buy its own GPU servers and run AI in-house, and the driver is client confidentiality

Financial Times; reported by Pulse 2.0 (Amit Chowdhry)

Latham & Watkins bought Nvidia GPU servers and is fine-tuning an open-weight model on infrastructure only its own people can reach, so that the most sensitive client data need not go to an outside cloud or AI vendor. The firm’s CIO told the FT some information is sensitive enough that it does not want it with “any cloud vendor.” Latham is keeping commercial tools too, and is building the ability to route a task to an in-house model or a third-party service depending on sensitivity, which it frames as a hedge against vendor pricing, access, and product changes. The $8.3 billion-revenue scale is out of reach for most firms, but the underlying question of where the client’s data gets processed is one every firm should address in its AI vendor diligence.

OpenAI launches “Astra for Law,” putting a frontier model and its own case-law index against the incumbent legal-research stack

LawSites (Bob Ambrogi), with OpenAI’s announcement

On September 17, OpenAI introduced Astra for Law, its GPT-6 Astra model configured for legal work with a built-in search index of U.S. case law, statutes, regulations, and administrative decisions — built in part on the Free Law Project’s CourtListener collection — plus custom instructions for legal analysis and drafting. It reaches selected firms first through a “Trusted Access” program in ChatGPT and Codex, with API access to follow, and pairs the model with legal-grade controls: Zero Data Retention on the API for eligible firms, ChatGPT Enterprise usage excluded from human review by default, and a governance collaboration with Latham & Watkins on ethical walls and permissions. For members, two things matter: those controls are becoming the yardstick for any AI vendor handling client-confidential prompts, and OpenAI is now a direct competitor to the established legal research and drafting tools rather than a general-purpose layer beneath them. The accuracy figures OpenAI cites are its own, and its launch page benchmarks against a competing model, so read them as vendor-reported; a legal-grade index still does not relieve the duty to verify, as the expert-report item elsewhere in this issue illustrates.

Florida’s local data-center moratorium wave reaches Palm Beach and Orange counties

ClickOrlando/News 6, with Orlando Weekly

The local pushback against AI data centers in Florida keeps spreading. Palm Beach County advanced a one-year moratorium on new hyperscale data centers, giving first-reading approval on August 27, with a second reading and possible adoption set for September 24. Orange County commissioners signaled support on September 15 and directed staff to study developing one, both citing water and power strain. This sits on top of the zoning authority SB 484 preserved for local governments when the state cost-shifting law took effect July 1: Issue No. 3 covered the Public Service Commission side of that law, and this is the county side playing out. Any client with a siting, land-use, or utility stake in a Florida data center now has to watch the county commission as closely as the PSC.

The Artificial Intelligence Task Force welcomes participation from all Business Law Section members. Attend a meeting, suggest an article, or help shape the Section’s work in this area. Visit the Task Force page or contact any Task Force officer.

The BLS AI Advantage is edited by the Artificial Intelligence Task Force officers: James Moon, Chair (jmoon@melandbudwick.com); Valerie David, Co-Vice Chair (vdavid@mgfirm.com); and Michael Bittner, Co-Vice Chair (mike@bittnerlegalgroup.com).