An AI now reads your contracts before you do, and it decides which ones you even need to look at. A Delaware firm's president had to personally apologize to a judge this week, and your malpractice carrier is quietly rewriting the rules for who pays when that kind of thing happens again. This issue is about who's actually doing the work now, and who answers for it when it's wrong.
THE LEAD PLAY
The First Read on Every Contract Is No Longer Yours
Spellbook rolled out Autonomous Contract Management on June 30. The system pulls incoming contracts straight from your email, Slack, and Salesforce, reviews and redlines each one against your standards, and sorts them into "ready to approve" or "needs a closer look," before a lawyer ever opens the document. CEO Scott Stevenson's pitch is blunt: lawyers wake up to a queue that's already been handled. A feature called Radar launches later this year to flag when a regulatory change makes a clause in your existing contract library obsolete. More than 4,500 legal teams already use the platform, including Dropbox's in-house group and the firm Kennedys.
Spellbook's own research team pulled a sample of SEC-filed contracts and found drafting errors in 60% of them. About one in 40 carried a high-risk error. That's the baseline for the current process, the one where a human reads the contract first. The company is betting that AI reading it first produces fewer errors, not more. It might. But the lawyer isn't reviewing the contract anymore. The lawyer is reviewing the AI's opinion of the contract, and that's a much easier thing to skim past.
Ironclad, DocuSign, and most of the rest of the contract lifecycle field are building the same kind of agentic layer. This isn't a one-vendor story. It's where the category is headed: from storing and routing contracts to deciding what happens to them.
The Play this week: If you're evaluating or already running an agentic contract tool, write down an explicit line: which contract types can be triaged and redlined without a human seeing the original document first, and which types require a lawyer to read the raw contract before they see the AI's version of it. High-volume, low-risk paper (NDAs, standard vendor terms) is a reasonable place to let the AI go first. Anything with unusual liability language, a new counterparty, or a deal-specific carve-out is not. Set that line yourself. Don't let the default settings set it for you.
SUPPORTING PLAY 1
Nobody's Asked What Your Malpractice Policy Actually Covers
Richards Layton had until today to explain to the Delaware Court of Chancery why it and one of its directors shouldn't be sanctioned over a brief containing hallucinated case law. The sanctions request itself isn't new territory; courts have been doing this since Issue #1. What's different here is who had to answer for it. Richard Rollo, the attorney who signed the brief, filed an affidavit. So did Paul Heath, the firm's president, who had no direct role in the underlying case. When an AI failure becomes the firm's problem instead of one lawyer's problem, the exposure changes shape, and your malpractice carrier is already thinking about that shift even if you aren't.
A parallel case closed out in Tennessee this week. Baker Donelson collected more than $45,000 in fees from a plaintiff's firm whose filings against it relied on unverifiable AI-generated citations. Different mechanism (Rule 11 fee-shifting instead of Chancery sanctions), same underlying exposure: bad AI output in a filing costs someone real money, and it's rarely the AI vendor.
WTW's latest Insurance Marketplace Realities report calls the past year a structural break in the professional liability market, and that's the part that should worry you more than either case alone. No major carrier has attached a public, named AI exclusion to a standard lawyers' professional liability form yet. But underwriters are asking every renewal applicant whether the firm uses AI, and ALAS, the mutual insurer covering many of the country's largest firms, has already told policyholders in writing that generative AI use could trigger a claim their professional liability policy might not cover.
The Play this week: Call your broker. Don't wait for renewal season. Ask directly whether your current policy responds to a claim involving an AI-generated hallucination or a missed clause, and whether a technology-failure exclusion could be invoked to deny it. Get the answer in writing. If it's vague, or if the answer depends on facts nobody at your firm has documented, that's what you bring to whoever owns the firm's insurance relationship, before a claim forces the question instead of you.
SUPPORTING PLAY 2
Reed Smith Put One Name on Every AI-Adjacent Team It Had
Reed Smith folded four previously separate groups, its staff attorney program, its RED discovery unit, Leeds Global Solutions, and Gravity Stack, into a single business unit called Reed Smith Legal Solutions. One leadership structure, one set of numbers to report, covering discovery, investigations, due diligence, contract management, legal ops support, risk management, and high-volume disputes. The firm's global managing partner framed the move around clients wanting one partner across the full lifecycle of legal work instead of four separate vendors operating inside the same firm. The unit already has at least one live product: Aquarius, an automated legal services platform built for crypto-asset issuers.
You don't need Reed Smith's headcount to learn from this. Most mid-size firms and legal departments run AI initiatives the way Reed Smith used to: e-discovery has its own tool and its own metrics, contract ops has its own tool and its own metrics, and legal ops runs a dashboard nobody else on the team sees. Issue #6 flagged that 93% of legal teams can't show whether their AI spending paid off. Fragmented ownership is a big part of why. When three people can each point to a different pilot and nobody owns the total picture, the question "did this actually work" doesn't have anyone assigned to answer it either.
The Play this week: List every AI-touched workflow currently running in your firm or department: e-discovery, contract review, intake, research, whatever's live. For each one, write down who owns it and who reports on whether it's working. If the honest answer is three different people acting informally, that's the gap. You don't need a new business unit to close it. You need one name next to "owns AI ROI reporting," even if it's a part-time addition to a role someone already has.
QUICK HITS
Arnold & Porter and Pillsbury took opposite approaches to the same hire this week. Arnold & Porter promoted its director of IT applications into the newly created Chief AI Officer role. Pillsbury went outside and hired AI consultant Oz Benamram as its first CAIO. Issue #5 argued the internal-promotion path is the one that scales for firms that can't win a $440,000 bidding war. Two firms just ran both experiments at once. Worth watching which one has an easier first year.
Epiq acquired flexible legal talent provider Tenor Legal. Another sign that legal staffing and AI-enabled delivery are consolidating into the same companies instead of staying separate markets, the same direction Reed Smith just moved in from the inside.
Apple sued OpenAI over alleged trade secret theft, accusing a former Apple hardware executive of using confidential project details while recruiting for OpenAI's hardware team. Not a legal-tech story on its face, but worth flagging if your firm handles executive recruiting disputes or trade secret litigation touching AI companies poaching from hardware teams.
Percipient launched a free tool for running blind, head-to-head comparisons between AI models. If Issue #1's stack-mapping exercise left you without a clean way to test vendor claims against each other, this is a low-cost way to do it before your next renewal conversation.
That's Issue #7. Before you approve the next contract an AI already cleared, ask what it skipped, not just what it caught. See you in the next one.