The Trust Spread Has a Published Closing Schedule

Most of legal AI is carrying a position its holders describe as a moat and a trading desk would book as a spread: long enterprise pricing, short the frontier labs' trustworthiness, with the gap between the two booked as gross margin. The wrapper charges enterprise rates because a law firm's IT committee cannot yet trust the raw model with its data, its confidentiality, its compliance posture, and the wrapper sells that trust back at a markup. Mark the position at entry: August 28, 2023, OpenAI launches ChatGPT Enterprise with SOC 2 compliance, encryption in transit and at rest, and one load-bearing sentence, "We do not train on your business data or conversations." Mark it at present: June 3, 2026, Shawn Curran, CEO of Jylo, quoted in Artificial Lawyer's piece on legal AI's token price problem: "firms paid a premium for wrappers when the supply chain models were useless."

Thirty-three months between the marks, and the verb at the exit is in the past tense, from a participant.

That the spread compressed is not this memo's claim; by June 2026 it is barely news. The claim arrives here, in the third paragraph: the spread did not leak away by accident. The closing schedule was published, dated, and signed by the counterparty · every certification, every residency region, every audit-log feature a public, dated haircut to a specific line in somebody's pitch deck, on the record since March 2023. Legal AI's dominant business model has been a trade against a counterparty that announces its deliveries in advance. This memo prints the tape, names who is still carrying the position under other names, and identifies the one asset in the legal stack marked against the model's floor instead of its ceiling.

1 · Name the trade

Call the gap what it is. The trust spread is the rentable distance between a frontier model's capability ceiling, what the model can do, and an enterprise's trust floor, what the enterprise will permit it to do. A wrapper's revenue line lives inside that distance: the model could already draft the clause, summarize the data room, run the privilege screen, but the firm could not yet accept the raw model's terms, so the wrapper interposed itself and priced the interposition.

The defining property of a spread is that it is rented, never owned. Both of its legs move on someone else's schedule, and in this trade one counterparty controls both: the labs raise the capability ceiling with every release and raise the enterprise trust floor with every certification, and the wrapper's margin is what remains between them. One piece of marking vocabulary, used throughout, sorts the whole landscape into two piles. An asset is marked to the ceiling when its revenue lives in what the model cannot yet be trusted or permitted to do; it compresses on the labs' schedule. An asset is marked to the floor when its cost base lives in what the model can already do; it expands on the same schedule. Everything below is an application of that one distinction.

If it is a spread, there is a tape.

2 · The tape · one leg marked in public, the other marked once, at exit

A spread needs two marks, so honesty about this table comes before the table. The lab leg prints in public: launches, certifications, attestations, dated by the counterparty itself. The wrapper leg structurally does not print. Enterprise contract values, renewal discounts, and churn are private; the repricing happens in silence, renewal by renewal, and surfaces only when a participant confesses. Curran's past tense is that confession, and it is the only wrapper-side mark this memo claims: one mark, at exit, from a CEO inside the trade. A single exit mark suffices for the same reason a single settlement print suffices on an expired contract, and the asymmetry is itself part of the finding: one side of this trade publishes its closing schedule; the other side's losses are booked in the dark. The table accordingly claims one leg, and says so in its title.

The Closing Schedule · the lab leg of the trade, 2023-2026

Exit mark, the wrapper leg's single print: June 3, 2026 · Curran, in Artificial Lawyer: firms "paid a premium for wrappers when the supply chain models were useless."

¹ Zero Data Retention exists as an approval-gated current offering for eligible endpoints; its start date is not publicly recorded, so it is cited as a current offering, not as a dated milestone. Both labs' SOC 2 Type 2 attestations are current; first-issuance dates are likewise not public, which is why neither appears as a row. ² Anthropic's own European residency is reported only in secondary sources and is not hard-dated; it is excluded from the table rather than soft-dated into it. ³ The one row in this table dated to the month rather than the day: coverage clusters in the first week of February 2026, with the LawSites piece this row quotes running February 5, and no day-precise release date is published; we flag the imprecision rather than manufacture precision.

Read the tape's shape. Six of the nine rows are the labs deleting trust premiums: security, audit, governance, residency, regulated data. The last three rows are something else: the labs entering legal itself, and the final two entries, May 12 and June 1, sit twenty days apart. Both labs moved on the legal vertical inside three weeks. The schedule's last phase is not the counterparty closing your spread. It is the counterparty taking your seat.

3 · Sequoia's autopilot, by name

The most sophisticated answer to this tape was published before its final rows printed. Julien Bek's "Services: The New Software" (Sequoia, March 5, 2026) states the escape route in two clean sentences: "If you sell the tool, you're in a race against the model. But if you sell the work, every improvement in the model makes your service faster, cheaper, and harder to compete with." And the slogan: "A copilot sells the tool. An autopilot sells the work." On this thesis the next trillion-dollar company is "a software company masquerading as a services firm," and the legal names attached are Harvey and Crosby. Bek nowhere uses the words wrapper or trust, and we will not put them in his mouth. The attack is on the logic, not the vocabulary.

The logic holds on exactly one condition: that the lab does not also sell the work. Rows eight and nine of the tape are that condition failing in public. A lab shipping twelve practice-area plugins and twenty-plus connectors into legal's systems of record, available to every paid plan, and a lab hiring Ironclad's co-founder to lead a legal vertical, are labs selling the work, or close enough that the difference is a product cycle. And when the lab sells the work, "selling the work" without owning a practice is not an exit from the spread trade. It is the same short position re-marked against a higher reference price. Sarah Tavel supplied the pricing logic back in August 2023, two and a half years before the Sequoia piece: work, unlike software, is "priced relative to the cost of a human performing the work instead of as a productivity improver." That is the appeal, and that is the exposure. The copilot was short the model's trustworthiness against software pricing. The autopilot is short the model's capability against human pricing, the widest spread on the board, which makes its margin the largest single target on the tape. Selling the work does not escape the race against the model. It raises the stakes of losing it.

4 · The healthcare transposition

One other licensed American profession ran a version of this experiment under a corporate-practice bar, and the argument here is structural, labeled as such, with no dated historical claims attached. Where medicine's corporate-practice doctrine barred outside ownership of the practice, outside capital's durable route into practice economics was the management services organization beneath the practice, not any particular vendor position above it. The obvious counterexamples mark the claim's edges rather than refute it: Epic is a durable vendor above the practice, and what it owns is a system of record, not a practice's economics; Optum, the services arm of the largest payer, bought practice economics outright, a route that exists in medicine's patchwork of corporate-practice rules and has no analog behind Rule 5.4. Neither is a trust-premium vendor that survived its premium. The structure that held practice economics under the bar was the MSO, current enough on the AI question to carry compliance literature of its own: Nixon Peabody and Stout's June 2026 analysis of AI-enabled MSO services treats fair-market-value and fee-splitting risk for exactly this entity, in exactly this position. Where the practice cannot be owned, capital's durable seat is beneath it.

5 · The survivor categories also compress

The standard rebuttal to wrapper pessimism names two survivor categories, and both are product positions marked to the ceiling. The first is the workflow specialist: the wrapper that survives on integrations, practice-specific tooling, and embedded process. Row eight of the tape prices that category directly: on May 12 a lab shipped the category's entire roadmap, free with the subscription. Jason Lemkin had already described the underlying physics from the SaaS side: "your prompt layer is portable". Workflow thickness made of software is thickness the counterparty can ship, and the tape shows it shipping.

The second category is the data fortress: the vendor whose corpus of matters compounds into a moat. That position dies on ground this series already holds: why the matter corpus does not lawfully pool, and where legal's data floor actually sits, is the subject of "Privilege Is a Zoning Law for Data," cited in this one line and left there.

6 · The floor is forced, not chosen

If product positions are marked to the ceiling, the obvious trade is to own the floor directly: own the practice. A serious cohort is executing exactly that, and it deserves engagement on its real strength. Covenant is an AI law firm built for private-market investors, reviewing LPAs in two business days for clients collectively managing over $200 billion. Crosby is an actual law firm, Sequoia-backed at seed and through a $60 million Series B, negotiating MSAs, DPAs, and NDAs with lawyers and in-house AI · Sequoia's own autopilot example holds the floor by being the practice. Eudia raised a $105 million Series A from General Catalyst and bought ALSPs, Johnson Hana and Out-House, to serve Fortune 500 legal departments. All three are genuinely in complex work, conceded without hedging. All three hold the floor by being or owning the practice.

And that is the position's boundary: in 48 American states, Rule 5.4 says outside capital cannot replicate it.⁴ Jonathan Molot put the corporate-finance consequence in a sentence: "Law firms simply do not have permanent equity," with Rule 5.4(d)'s bar on nonlawyer ownership as the premise he builds on. The own-the-firm thesis works wherever you can own the firm, which at national scale, for outside capital, is approximately nowhere. So the floor position in US legal is not a strategy anyone chose from a menu. It is structurally forced into the one entity capital may own that sits beneath the practice: the MSO.

Here is this section's claim, and the sentence this memo exists to print: every improvement in the model compresses the wrapper's revenue and compresses the MSO's cost base. Same event, opposite mark. The wrapper's revenue lives in the gap between ceiling and floor, so each lab release and each certification deletes a slice of what it can charge for. The MSO's economics live in operating a practice's business layer, so each identical release deletes a slice of what its operations cost. Section 2's schedule reads, from the wrapper's side, as a list of haircuts; from the MSO's side, as a list of cost reductions delivered free by what is plausibly the most heavily capitalized R&D effort in commercial history. One position dreads the labs' roadmap; the other compounds on it. Who captures that cost-base compression, the firm or the MSO, is a fee-clause question, treated at length in "The AI Dividend Has an Address," and handed off in this line.

⁴ Counting convention, so the house agrees with itself: the bar holds in 48 states, with the Arizona ABS regime and the Utah sandbox as the carve-outs, plus the District of Columbia's narrower allowance for minority nonlawyer partners · the formulation this series uses throughout. Commentary that says 49 is treating Utah's sandbox as a pilot rather than a repeal. Nothing in this memo turns on the difference.

7 · The thicker application layer, answered

The last serious rebuttal comes from the application-layer thesis: the venture position, argued most prominently out of a16z, that in every platform wave value accretes to applications built thickly on top of commodity models, so the answer to lab encroachment is more application, deeper workflow, richer integration, until the model underneath you commoditizes in your favor. We state the position generally because we are answering the position, not a quotation.

In most software markets the thesis has history behind it. Legal is the limiting case that breaks it, for one reason visible on the tape: thickness made of software is thickness the lab can ship, and ship free. The May 12 release was not a model update; it was application thickness, connectors and practice plugins, delivered as a bundle at a marginal price of zero. The June 1 hire bought application-layer judgment outright. When the platform owner manufactures your thickness as a loss leader, thickness is not a moat; it is a roadmap item. The only thickness the lab cannot ship sits behind a license: practice operations, supervision structures, accountable delivery of legal work under a bar number. So the durable application layer in legal is the one built from operations rather than software, attached to a practice it cannot own, with an MSA underneath it. Which is not an application in the lab's sense at all. It is the entity the rest of this memo has been describing.

8 · The procurement terminal state

Walk to the end of the closing schedule, because the end state is not abstract; it is a spreadsheet. The firm's vendor stack at terminal state is a procurement worksheet with a row per tool: research platform, drafting assistant, e-discovery suite, DMS, the lab subscription itself. Each row carries the same fields: vendor, seats, per-seat price, renewal date, and the column procurement adds the moment a capability commoditizes: alternative. The tape's last phase fills that column in. When a lab ships practice plugins and twenty connectors free with the subscription the firm already pays for, every wrapper row acquires a permanent zero-dollar comparable one line below it at every renewal. That is what a closed spread looks like in an operating document: not bankruptcy, a benchmark. The wrapper's terminal customer is not the practice; it is procurement, whose job is to make every row replaceable by the row beneath it.

Now ask who owns the worksheet. Vendor selection, contract negotiation, renewal benchmarking, the annual exercise of playing the zero-dollar comparable against the incumbent: that is a management function, and in a practice that has separated its business layer from its professional layer, it sits inside the MSO by contract. The MSO is not a row in the spreadsheet, and the reason is the instrument: a vendor holds a license into the firm's stack, while the MSO holds the MSA, which owns the systems and sits beneath the practice rather than among its suppliers, the reading "The MSA Is the Cap Table" gave that document at the start of this series. A firm can RFP any row in its stack. It cannot RFP the entity that conducts the RFP and owns the stack the rows plug into. That is the whole distance between the two end states, and it is the sentence this trade has been converging on since March 2023: the wrapper's end state is a line item in the firm's vendor stack; the MSO's end state is the stack's owner.

The practice pushes back

A lawyer thirty years into practice reads all of this and objects, and the objection deserves its full strength:

"Your spread talk mistakes which trust matters. No client has ever hired a SOC 2 report; they hire a lawyer who answers for the work with a license, a malpractice policy, and a name on the signature block. ABA Formal Opinion 512 puts AI squarely under Rules 5.1 and 5.3: supervising lawyers must review AI-assisted work product, with the required verification depending on the tool and the task. California's guidance requires me to 'critically review, validate, and correct both the input and the output,' and this spring the California bar proposed hardening verification into rule. The labs shipping connectors changes nothing about who gets sued. The trust gap you say is closing is the enterprise-IT gap; the professional-liability gap never closes, and it belongs to me. And your economics are no better than your doctrine: even if your MSO owns the cost base, its fee must sit at fair market value, and FMV re-benchmarking hands the savings back to my firm at every reset. Your floor is a term annuity, not ownership. Strip the vocabulary and you have described a vendor with a longer contract."

Concede the core, because it is correct and it is load-bearing. The liability floor belongs to the lawyer, permanently. Opinion 512 and the California guidance are not obstacles to this memo's thesis; they are its mechanism. The review duty is the precise reason the labs' closing schedule stops at the license line: the one premium with no row on the tape, the one no certification can haircut, is the lawyer's professional accountability, because it is not a trust deficiency the lab can remediate. It is an allocation of liability the profession will not let leave the profession. Concede also that the review tax bounds the savings; supervision is a real cost that automation does not delete.

But notice what the objection actually protects. The supervision duty makes the lawyer non-disintermediable. It says nothing about any entity beneath the lawyer, which is why "a vendor with a longer contract" is the right challenge and the procurement section is its answer. A vendor is swappable because it holds a license and faces a benchmark every renewal; the MSO is not swappable by that machinery because it is the machinery, owning the systems under the MSA and conducting the procurement that disciplines every vendor row · the position section 8 walked. Swapping a vendor is an RFP. Swapping the MSO is unwinding the instrument that holds the firm's entire business layer, which is not procurement; it is divorce. And the FMV jab is conceded as the right objection in the right place; how long the dividend lasts and what resets it is the ground "The AI Dividend Has an Address" already walked. The honest summary of both concessions: wrapper and MSO both face compression, and only one survives it as counterparty rather than casualty. A fee that reprices is a different asset class from a vendor row that gets replaced.

The two marks

Strip the memo to its instruction. When a legal AI deck says moat, ask one question: which mark is the asset carried at? If revenue lives in what the model cannot yet be trusted or permitted to do, the asset is marked to the ceiling, and the counterparty has published the schedule on which that mark declines; the table above is that schedule to date. If the cost base lives in what the model can already do, behind a practice that Rule 5.4 keeps unownable and above an MSA that makes the business layer ownable, the asset is marked to the floor, and the same schedule compounds it.

Thirty-three months of tape, one exit confession, and a counterparty now hiring for the legal vertical. The trade was never a moat. The wrapper rents the gap; the MSO owns the floor.

This memo is published by Jopese, a legal management services organization operated by HIRO PARTNERS LLC, a Texas limited liability company. It is offered for educational and analytical purposes only. It is not legal, tax, or investment advice, and it is not an offer to sell or a solicitation of an offer to buy any security or service. Jopese is not a law firm and does not provide legal advice or legal services; legal services are delivered by an independent law firm under a separate engagement in which Jopese does not participate. References to specific companies, funds, transactions, products, ethics opinions, and regulatory developments are drawn from public sources and are provided as market commentary, not as an endorsement, a recommendation, or a representation of any relationship. Characterizations of trading positions, spreads, and marks are analytical metaphors, not descriptions of any security or investment position.