The Innovation Attorney Podcast

The Innovation Attorney

The Innovation Attorney is a publication about the intersection of venture capital, technology, public policy and law. theinnovationattorney.substack.com

  1. 13h ago

    AI Usage Based Pricing Contract Renewal Terms

    A venture backed company that renews its software contracts on the old per seat template this year is agreeing to pricing mechanics it cannot audit, cap, or benchmark, because the vendor on the other side of the table has already rebuilt its own billing engine around usage and outcomes. Salesforce sold its Agentforce product under a flat $2.00 per conversation fee at general availability on October 29, 2024. Seven months later, on May 15, 2025, it added Flex Credits, priced at $500.00 per 100,000 credits with a standard production action metered at $0.10. Three months after that, on August 19, 2025, it layered in pay as you go and pre commit payment options, and by October 2025 it was offering a negotiated Agentic Enterprise License Agreement to its largest accounts. Four billing constructs, inside thirteen months, from a single vendor. KeyBanc Capital Markets later reported that Agentforce deal volume grew 3.7 times over that stretch while the share of deals converting to paid usage actually fell, from roughly 60 percent to roughly 51 percent. A pricing engine that unstable is not a glitch: it is what PYMNTS described in February 2026 as the break between seat count and software revenue that artificial intelligence has opened across the industry, and it is the environment every SaaS customer is now renewing into. The fix starts with the metering unit itself. A usage cap paired with a true up window, rather than real time overage billing, lets a customer see consumption against a contracted ceiling and settle the difference once, at the contracted rate, rather than watching an invoice move with every token processed. Salesforce priced Agentforce at $2.00 per conversation at general availability on October 29, 2024, then added Flex Credits at $500.00 per 100,000 credits on May 15, 2025: a founder negotiating a renewal should ask the vendor to publish the same kind of per unit rate before signature, not after the first invoice. Audit rights need the same rebuild. A clause written for a flat license assumes the vendor is checking a headcount against a key. Usage based billing turns that clause into the vendor’s right to inspect the telemetry stream that generates its own invoice, so the customer should cap audits at once every twelve months, require thirty days of notice, and confine any audit to vendor furnished usage reports rather than direct system access. The same obligation runs in reverse for any company embedding a vendor’s AI feature into its own product: that company needs its own audit rights over the vendor’s model version and data handling, a protection most vendor paper omits entirely. Most favored nation clauses need to widen too. A clause that only compares published per seat list price will never trigger against a vendor selling four different constructs to four different customer segments, the way Salesforce did between October 2024 and October 2025. The fix is to define the comparison set as any pricing construct offered to a similarly situated customer, measured by total cost per unit of consumption, not sticker price, and to keep the clause alive for the full term rather than letting it lapse at the first renewal. The regulatory exposure runs through the company’s own downstream contracts as much as through the vendor’s. The United States Court of Appeals for the Eighth Circuit vacated the Federal Trade Commission’s 2024 Click to Cancel Rule in 2025 for skipping a required regulatory analysis, and the agency answered in March 2026 with an Advance Notice of Proposed Rulemaking to bring the rule back, with comments due April 13, 2026. In the meantime, the Federal Trade Commission has kept enforcing the Restore Online Shoppers Confidence Act directly, a statute that produced a $2.5 billion settlement against Amazon and an $8.5 million settlement against Care.com over cancellation and disclosure practices. A company that reprices its own customer subscriptions around a per resolution or per conversation fee inherits that exposure the moment it bills a customer without clear disclosure, consent, and an easy cancellation path, regardless of whether the 2024 rule is ever reinstated. Pricing benchmarks exist now, so a founder does not have to negotiate blind. Intercom’s support agent, Fin, charges $0.99 per resolution and has crossed $100 million in annual recurring revenue on 393 percent annualized growth. Zendesk charges between $1.50 and $2.00 per automated resolution. Sierra AI structures outcome based contracts starting around $150,000.00 annually and charges nothing when a case escalates to a human agent. A vendor proposing a materially higher figure for a comparable unit of automated work should explain the gap, and a founder with that range in hand has standing to ask for it. None of this argues for refusing usage based or outcome based pricing. It argues for renewing into it with the same discipline a founder would bring to a seat count: a capped unit rate, a bounded audit, a most favored nation clause that actually reaches the vendor’s current product, and a downstream subscription disclosure that survives a return of the Click to Cancel Rule. The next piece in this series looks at who owns the usage and outcome telemetry a billing engine generates once the contract ends, and what it takes to write a data rights clause that turns a vendor’s own metering data into the customer’s proprietary training set. Interested in analysis about the intersection of venture capital, tech, artificial intelligence, public policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/ This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit theinnovationattorney.substack.com/subscribe

  2. 1d ago

    Employee Equity Liquidity After Carta And Pulley Exit Secondaries

    Startups now design employee equity liquidity through company sponsored tender offers, broker matched secondary sales, direct buybacks, or option financing, since Carta exited secondaries and Pulley closes December 8, 2026. Carta exited the secondary trading business on January 9, 2024, citing a conflict of interest it could not manage between holding customer cap table data and trading against it, and Pulley, the rival platform that built its business on being the alternative to Carta, will shut down entirely on December 8, 2026, sending its customers back to Carta for cap table administration only. The two platforms that employees and founders used to associate with liquidity have both stepped back from the part of the job that actually moves cash to a vested employee. The mechanism most founders overlook is not the tax bill or the paperwork. It is that removing the platform does not remove the demand. Employees at any company approaching a meaningful valuation still want to convert years of vesting into a down payment on a house or a paid off student loan before the next round resets their strike price math, and if the company does not build a channel for that, employees build their own, informally, through a broker who calls without knowing whether the transfer requires board consent. Four structures now have to be assembled deliberately, since neither platform still offers this as a feature to switch on. The first, and the one boards already recognize, is the company sponsored tender offer: a written offer to purchase vested shares, held open at least twenty business days under the Securities and Exchange Commission’s tender offer rule for unregistered securities, priced at a premium over the current 409A common valuation, and closed on a pro rata basis if oversubscribed. Databricks ran one alongside its $10 billion Series J round in April 2025 at a $62 billion valuation, and the pattern that emerges from deals like it is that the clearing price sits between the old 409A number and the new preferred price, and the 409A that gets recalculated afterward resets somewhere in between instead of jumping straight to the new figure. The second structure routes around the company through a licensed broker platform. Nasdaq Private Market announced a partnership with the secondary investor G Squared on January 8, 2026, built on what it describes as nearly $70 billion in transaction volume across more than 875 company sponsored programs, a scale that suggests this channel is consolidating around a small number of platforms now that a major cap table incumbent has twice declined to compete for the business. A special purpose vehicle typically holds the purchased shares on behalf of multiple outside buyers and counts, for now, as a single record holder under Section 12(g) of the Exchange Act, which matters because a company that crosses 2,000 record holders has to begin filing with the Securities and Exchange Commission whether or not it wants to be a reporting company yet. The third and fourth structures skip the outside buyer altogether. A direct buyback funded from the company’s own cash keeps the cap table simple but consumes runway dollar for dollar and asks incoming investors at the next round to trust that the balance sheet spent on cap table cleanup was not spent instead on the growth they are pricing. A nonrecourse financing arrangement takes the opposite path, advancing an employee the cost of exercising options and the associated tax liability in exchange for a share of eventual proceeds, so the employee owns shares without a sale ever happening and without writing a personal check, though the financing firm only gets paid back if the company eventually delivers a real liquidity event. None of the four structures is free of the same clause that has killed secondary transactions since long before Carta or Pulley existed. Every venture financed company’s certificate of incorporation carries a right of first refusal and a board consent requirement on any transfer of common stock, and the deal that falls apart rarely falls apart on price. It falls apart because counsel did not calendar the refusal window against the broker’s closing date, the company exercises its right to match the offer or simply declines consent, and the employee is left holding illiquid shares and a tax bill from an option exercise made in anticipation of a sale that never closed. Facebook’s own experience between 2009 and 2012, trading through SecondMarket in volumes large enough to push it toward the pre Jumpstart Our Business Startups Act threshold of 500 record holders, is the clearest illustration that an uncontrolled secondary market does not stay a private matter between a buyer and a seller. It becomes a securities law event for the whole company, and in Facebook’s case that pressure was a documented factor in when the company had to go public. There is a tax clock running underneath all of this that most tender offer memos treat as an afterthought. An employee who exercises an incentive stock option and holds the shares for five years can exclude a substantial share of the gain from federal tax under Section 1202 of the Internal Revenue Code, and tax legislation that took effect on July 4, 2025, replaced the old all or nothing five year rule with a tiered schedule that pays out 50 percent after three years and 75 percent after four. Selling into a company’s first tender offer at year three instead of waiting two more years for the full exclusion is a real dollar decision, and it belongs in the offer materials the employee reads before the window closes, not in a conversation with a tax preparer the following April. The next piece in this sequence works through how a board should actually price a tender against the strike price reset it creates for every option granted afterward, using the gap between the Databricks and Anthropic 2025 tenders as the working model. Interested in analysis about the intersection of venture capital, tech, artificial intelligence, public policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/ This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit theinnovationattorney.substack.com/subscribe

  3. 2d ago

    Stock Option Backdating Cost the UnitedHealth CEO $468 Million

    Stock option backdating cost former UnitedHealth Group chief executive William McGuire $468 million on December 6, 2007, after the SEC alleged he selected low price grant dates retroactively. Backdating means writing a grant date earlier than the date the board acted, so the exercise price matches a historical low. A $30 exercise price on a stock that trades at $50 hands the executive $20 per share on the first day. The practice violated federal securities law when companies hid that gap from investors and auditors. The numbers explain the reaction. Professor Erik Lie studied 39,888 option grants at 7,774 companies from 1996 to 2005 and estimated that 14 percent of the grants were backdated or manipulated. Roughly 200 companies faced federal investigation or restated earnings by January 2008. The research and a March 2006 newspaper investigation produced the caseload, and the newspaper series won the 2007 Pulitzer Prize for Public Service. How did companies profit from backdating without recording the cost? Accounting Principles Board Opinion No. 25 let a company record zero compensation expense for an option priced at market on the grant date, and required expense for an option granted in the money. Backdating relabeled in the money options as at the money options, and the expense never reached the income statement. The SEC alleged that Brocade Communications avoided reporting hundreds of millions of dollars of expense between 1999 and 2004, in part by dating grants for new hires to days before they had interviewed. For UnitedHealth the SEC alleged an understatement of $1.526 billion in stock based compensation from 1994 to 2005. FASB Statement No. 123(R), effective June 15, 2005 for most public companies, ended the arbitrage by requiring expense for all share based payments. It was issued in December 2004, more than a year before the press investigation, so the accounting incentive had already disappeared when the scandal broke. What did backdating do to taxes and proxy statements? Section 162(m) of the Internal Revenue Code capped the corporate deduction for executive pay at $1 million and exempted performance based pay. In Memorandum AM2009006, dated July 17, 2009, the IRS Chief Counsel concluded that a discounted option cannot qualify for that exemption, even if the exercise price is raised later. The same memorandum treated a backdated option meant to be an incentive stock option as a nonstatutory option, so the employee owes ordinary income tax on the spread at exercise. Proxy statements added a securities law problem. They told shareholders that exercise prices equaled fair market value on the grant date, and for a backdated grant that sentence was false. That placed the practice under Section 10(b) of the Securities Exchange Act of 1934 for fraud, Section 14(a) for the proxy, and Section 13(b)(2) for books and records and internal accounting controls. Did the courts and prosecutors treat backdating as fraud? The Delaware Court of Chancery answered for directors in Ryan v. Gifford on February 6, 2007: directors who approve backdated options in violation of a shareholder approved plan, while representing compliance, act in bad faith and breach the duty of loyalty. Section 102(b)(7) of the Delaware General Corporation Law does not exculpate that conduct. The same day the court decided In re Tyson Foods and allowed spring loading claims to proceed, which concern grants made days before good news and carry no false date. Prosecutors had a harder path. A jury convicted Brocade chief executive Gregory Reyes on August 11, 2007, and the Ninth Circuit reversed in United States v. Reyes on August 18, 2009 because of prosecutorial misconduct. A second jury convicted him on nine of ten counts on March 26, 2010, and the Ninth Circuit affirmed on October 13, 2011. The McGuire settlement carried a ten year bar from serving as an officer or director, and about $448 million of the $468 million was reimbursement under Section 304 of the Sarbanes Oxley Act. What changed in the reporting rules after the scandal? Congress had already acted. Section 403 of the Sarbanes Oxley Act of 2002 amended Section 16(a) of the Exchange Act to require a Form 4 within two business days of an option grant, effective August 29, 2002. Before that, an insider could report a grant as late as 45 days after fiscal year end, which left months to look back for a low price. Lie and Randall Heron found that backdating of unscheduled at the money grants fell from 23 percent to 10 percent after the rule, so the practice shrank and continued. The SEC then added disclosure. Its 2006 executive compensation rules require a table of grant dates, exercise prices, and grant date fair values, and a discussion of how the company chooses grant dates. Item 402(x) of Regulation S K, adopted December 14, 2022, requires a narrative of grant timing policies and a table for options granted to named executive officers within four business days before or one business day after a periodic report or a current report that discloses material nonpublic information. What should a startup board record on the day it grants options? A private company files no Form 4 and issues no proxy, so no outside clock catches a misdated grant. In my practice advising venture backed companies, the recurring defect is an exercise price taken from an expired valuation. A founder who asks whether a new hire can receive last quarter’s lower price is asking the backdating question in miniature. Section 422(b)(4) and Section 409A both tie the price to fair market value on the grant date, and a consent signed after its stated date does not move the grant date backward. The board resolution should state the number of shares, the exercise price or the formula for it, and the valuation relied upon, and every director should sign on or before the date the grant takes effect. Companies that follow that sequence in a board minute book pass the diligence review that investors run at Series A. The next piece in this series explains how a Section 409A valuation date and a board consent date together decide whether a venture backed company’s first option grant keeps incentive stock option status, including the specific consent drafting errors that cost founders that status. Interested in analysis about the intersection of tech, policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/ This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit theinnovationattorney.substack.com/subscribe

  4. 2d ago

    Deep Fission Gravity Reactor

    Deep Fission’s one mile deep Gravity reactor has no Nuclear Regulatory Commission license application on file as of September 2026, and its June 2026 prospectus describes the design as conceptual. The Commission’s pre application record under Docket 99902126 shows no date for a combined license submission. The Department of Energy approved a Nuclear Safety Design Agreement for the project in July 2026, a milestone that authorizes no operation. Gravity is a miniature pressurized water reactor fueled with low enriched uranium, which Deep Fission, a California company founded in 2023 and listed on Nasdaq since June 2026, intends to lower into a water filled borehole in hard rock. Data center developers and shareholders carry the exposure. Four other developers in the Department of Energy Reactor Pilot Program reached criticality on or before July 4, 2026. Deep Fission did not. The cost record is the weakest part of the file. The company removed its pre IPO guidance of $50 to $70 per megawatt hour from its prospectus, and rebuilding the calculation from the capital and operating figures it did disclose gives $289 for a first reactor and $148 for a repeat reactor. At June 30, 2026 the company held $93.0 million in cash, had used $42.5 million of it in six months, and carried a going concern warning. Two bodies of law frame the exposure. Atomic Energy Act Section 103 and 10 CFR Parts 52 and 53 govern the license the company still has to obtain, and an August 2024 NRC staff memorandum found no regulatory pathway for disposing of a reactor in a borehole. Securities Act Section 11 and Exchange Act Rule 10b5 govern the statements made to investors, and two plaintiff firms announced investigations on September 29 and October 1, 2026. Detailed Findings What is the Deep Fission Gravity reactor and how does it work The Gravity reactor is a four assembly pressurized water reactor, fueled with uranium oxide enriched below 5 percent uranium 235, that Deep Fission would lower into a cased borehole filled with water roughly one mile below the surface. The April 22, 2026 conceptual design description, revision 0, specifies four standard fuel assemblies in a 17 by 17 array of 264 rods each, a thermal rating of 25 megawatts, and a borehole of about 38 inches inner diameter with a nominal 2 inch casing. Three water loops move the heat: a primary loop that circulates by natural convection, a pumped secondary loop, and a tertiary loop that drives the surface turbine and generator. Containment is assigned to the fuel and cladding, a reactor canister, and the borehole itself, including its water column, a closure device, and the surrounding rock. The same document describes the design as being in a pre final phase, with the safety classification of structures and components preliminary and subject to refinement. The company attributes its cost and speed advantages to that geometry. The water column supplies roughly 160 atmospheres of hydrostatic pressure at depth, which the company says removes the need for the thick above ground containment structure and some of the safety systems a conventional plant carries. Spent fuel can be stored in the borehole or retrieved to the surface for conventional management. The company intends to take part in owning and operating its plants, which makes it the likely licensee and the party carrying nuclear liability. I stood reactor watches on a nuclear submarine, and the maintenance and casualty philosophy of that pressurized water plant rested on a trained crew being able to reach the equipment. A reactor a mile down a borehole removes that option from the design. Every component below the surface has to be qualified to run to retirement without a hand on it, and I did not find in the public record how that qualification will be demonstrated. How have the Gravity design parameters changed since 2024 The company’s own documents give different power ratings and borehole diameters for the same reactor. The 2024 white paper described a 30 megawatt thermal unit producing up to 10 megawatts electric in a 30 inch borehole. The April 2026 conceptual design description lists 25 megawatts thermal in a 38 inch borehole. The June 2026 prospectus describes boreholes of 30 to 50 inches, a first of a kind configuration of about 8 megawatts electric, and a repeat configuration of about 15 megawatts electric. The Commission’s pre application page lists 45 megawatts thermal and 15 megawatts electric at about 30 inches. A short seller report states that the chief executive told an August 2026 investor conference that the company is now targeting 5 megawatts electric in a 40 inch borehole and 15 megawatts electric in a 60 inch borehole, and that the chief executive acknowledged pushing the limits of borehole diameter. I did not locate a recording or transcript to verify those remarks. The direction matters because oil and gas and geothermal wells at depth are commonly drilled at diameters near 10 inches, as the same report notes, so each added inch of diameter moves the company further from proven drilling practice and adds drilling cost. Where does Deep Fission stand in the DOE Reactor Pilot Program Deep Fission has completed the second of four Department of Energy authorization milestones and has not reached criticality. The Department selected the company in August 2025 as one of eleven eligible projects, and the Other Transaction Agreement for reactor authorization took effect on November 14, 2025. The 81 page Nuclear Safety Design Agreement was approved in July 2026 and published in September 2026 with minimal security redactions, according to local reporting, and the short seller report dates the company’s announcement of the approval to early August 2026. The remaining milestones are the preliminary documented safety analysis and the documented safety analysis, and the last of those enables operation. Four developers reached criticality on or before July 4, 2026: Antares Nuclear with the Mark 0, Valar Atomics with the Ward 250, Deployable Energy with the Unity, and Aalo Atomics with the Aalo X. The short seller report states that Deep Fission targeted criticality by July 2026 in a September 2025 registration statement, that its chief executive described July 4 as the target on a November 2025 podcast, and that the company later dropped the target and characterized the milestone as a science experiment. Program status coverage in July 2026 lists the company as aiming for full power operation in early 2027 and commercial deployment in late 2027 or early 2028, and reports a test well drilled to 6,000 feet. The prospectus states that Department authorization does not permit commercial operation and does not replace the requirement to obtain NRC licenses. A company spokesperson said the same of the Nuclear Safety Design Agreement in September 2026. How long will NRC licensing of a borehole reactor take The company plans to file for a commercial license in the first half of 2027, and the Commission has announced no review schedule because no application exists. The closest benchmark is a conventional design. NuScale’s uprated 77 megawatt electric module received a standard design approval on May 28, 2025, about 22 months after the Commission accepted the application in July 2023, and that review covered an above ground module. NuScale’s original 50 megawatt electric design took roughly six years from application to the January 2023 certification rule. The Commission finalized Part 53 on March 26, 2026, published it on March 30, 2026, and made it effective April 29, 2026. The optional rule is risk informed, performance based, and technology inclusive, and Commission officials project design approvals in 18 months or less at about half the cost of a Part 50 or Part 52 review, consistent with the 18 month limit in Executive Order 14300. The company’s schedule, as the short seller report reads its investor materials, implies a decision within about a year of filing. A filing in the first half of 2027 and an 18 month review puts a decision no earlier than late 2028. The prospectus describes a development and deployment timeline of one year for the first of a kind configuration and two years for the repeat configuration. The short seller report states that a corporate overview presentation from August 2026 advertises deployment in six months. The documents do not say whether the six months begin after site characterization and licensing are complete. What did NRC staff say about the borehole concept NRC staff told Deep Fission in an August 21, 2024 memorandum that the white paper gave no technical basis for treating the surrounding rock as a heat sink and no basis for a regulatory pathway to store or dispose of the reactor in the borehole. The memorandum, prepared by the project manager in the Division of New and Renewed Licenses, records several concerns that bear on cost and schedule. Staff noted that heat load was a dominant factor at Yucca Mountain and that the geology there behaved more like a blanket than a sink. Staff described the borehole as a breach in the natural barrier the company relies on for containment, a concern that multiplies with multiple boreholes. Staff expected heavy neutron leakage and uneven burn in the 2 by 2 core, with possible exceedance of peaking factor limits and activation of the surrounding rock. Brine rich groundwater was flagged as a challenge to material performance, seismic characterization was tied to 10 CFR Section 100.23, and staff wrote that severing of the borehole in an accident seems somewhat likely, with the risk assessment unclear. On the company’s claim that boreholes can be drilled within days and the reactor emplaced and operating within weeks of delivery, staff replied that the basis was unclear and that excavation and geologic investigation at depth for Yucca Mountain took a substantial amount of time, well beyond days. The memorandum addressed the 2024

  5. 2d ago

    The Lyft Settlement For Gig Worker Misclassification Risk

    On October 1, 2026, Lyft agreed to pay $272.5 million to settle a California Labor Commissioner lawsuit accusing the company of misclassifying drivers as independent contractors from April 6, 2016, through December 15, 2020. Eighty seven percent of that figure, $236.875 million, goes straight to drivers rather than to the state treasury or to outside counsel. The Labor Commissioner’s office filed the underlying case in August 2020, which means Lyft spent six years and a California Supreme Court decision on Proposition 22 before writing the check. A parallel suit against Uber, filed the same month on the same legal theory, is still working its way through an arbitration dispute at the California Court of Appeal. If you run a venture funded company with any contractor workforce, that six year gap between filing and payment is the part that should worry you, not the dollar figure. California did not invent a new standard to catch Lyft. Dynamex Operations West, Inc. v. Superior Court (2018) set the rule that every worker is presumed an employee unless the company proves the worker operates free of its control, performs work outside the company’s usual business, and runs an independently established trade of the same kind. The legislature wrote that presumption into Labor Code section 2775 through Assembly Bill 5 in 2019. Voters then carved app based rideshare and delivery drivers out of it through Proposition 22 in November 2020, trading the ABC test for a net earnings floor and a health stipend, and the California Supreme Court upheld that trade as constitutional in Castellanos v. State of California (2024). The carve out is specific to transportation network companies and delivery network companies. It does not cover a company that routes data labeling tasks, content moderation jobs, home repair gigs, or grocery fulfillment runs through an app and calls the workers contractors. Uber’s case shows why the Lyft number matters even to startups nowhere near the rideshare market. Uber moved to compel arbitration of the Labor Commissioner’s claims, the San Francisco Superior Court said no in September 2022, and the California Court of Appeal stayed the case in December 2022 while it sorts out whether a state agency suing in its prosecutorial capacity is bound by a driver’s individual arbitration agreement. Most venture backed companies write a mandatory arbitration clause into every contractor agreement they sign, on the theory that it keeps disputes out of court and off the public record. That theory has never been tested against a state labor agency that is not a party to the contract and does not need the worker’s consent to sue. Until the Court of Appeal rules, a startup cannot assume its arbitration clause does anything at all against the one plaintiff with the authority Lyft just paid $272.5 million to settle with. The theory has already left the rideshare industry. White v. Mercor.io Corporation, filed in the Northern District of Texas in May 2026, accuses the AI infrastructure company of misclassifying more than 30,000 contractors who supplied training data labor to customers that include OpenAI, Anthropic, and Meta, allegedly to avoid payroll tax and benefit costs. A separate complaint against Surge AI describes a data labeler who was told what hours to work, what tools to use, and how to complete each task, while being paid as a contractor. Scale AI faces its own dispute over Kenyan content reviewers paid as little as $2 an hour to review violent material with no mental health support and none of the protections an employee would receive. None of these companies run a transportation network. Each one built a labor pipeline on the same control pattern Dynamex was written to catch. In thirty years of counseling technology companies through financings and acquisitions, I have sat across the table more than once while a buyer’s counsel found, deep in diligence, that a chunk of the target’s workforce looked like employees wearing a contractor label. It never kills the deal outright. It resets it. The buyer prices the exposure into an escrow, the closing slips while the parties argue about how far back the liability runs, and the founders spend the weeks before close learning exactly how expensive years of casual classification decisions can get. Labor Code section 226 penalizes every inaccurate wage statement. Labor Code section 203 adds waiting time penalties for every day of unpaid wages after separation. Claims under the Private Attorneys General Act, Labor Code section 2698, stack civil penalties per pay period across the whole contractor roster, not per lawsuit. None of that shows up on a cap table until someone asks. Federal law is no steadier. The Department of Labor’s 2024 rule built a six factor economic reality test for who counts as an employee under the Fair Labor Standards Act, effective March 2024. In May 2025, the Department told its own investigators to stop applying that rule in open enforcement matters and go back to the older totality of the circumstances test instead, without repealing the 2024 rule through notice and comment rulemaking. A founder who built a compliance file around the 2024 rule’s six factors already has a defensible record in hand. One who never built that file is leaving a gap for the next labor agency complaint to find. Whether the California Court of Appeal lets a state agency bypass arbitration in the Uber case, and whether White v. Mercor.io Corporation turns the AI training data industry into the next rideshare, are the two developments that will shape how much this costs the next venture funded company that gets the call. My next piece follows the Mercor.io docket and the arbitration ruling in the Uber case as both move through the courts this year. Interested in analysis about the intersection of tech, policy and the law? Check out my Substack channel. https://theinnovationattorney.substack.com/ This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit theinnovationattorney.substack.com/subscribe

  6. 2d ago

    Government Action and Fraud in Six American Financial Debacles

    Executive Summary Government deregulation and supervisory forbearance caused or worsened six American financial crises between 1982 and 2008, and fraud enlarged the losses in each one. The Savings and Loan Crisis, the dotcom crash, the Enron collapse, the WorldCom fraud, the 2008 mortgage crash, and the 2005 to 2006 stock option backdating scandal each began with a government decision that widened the room for risk or loosened the supervision of it. Fraud then converted ordinary business failure into public loss. The Savings and Loan Crisis cost taxpayers about $124 billion after the Garn St Germain Act of 1982 let thrifts make commercial real estate loans while regulators lowered capital requirements and kept insolvent institutions open. The dotcom crash erased roughly $5 trillion in market value after investment bank research conflicts went unregulated. Enron relied on a 1992 accounting approval and a 2000 derivatives exemption. WorldCom overstated assets by more than $11 billion through line cost accounting. The 2008 crash followed a 2000 statute that barred regulation of over the counter derivatives, a 2004 capital rule that left investment banks at leverage near 30 to 1, and mortgage fraud that grew twentyfold in suspicious activity reports between 2003 and 2007. Option backdating grew from a tax rule and an accounting rule that rewarded options priced at market. The statutes, settlement figures, and convictions described here are settled. The weight each cause deserves is contested. Congress answered the thrift losses with the Financial Institutions Reform, Recovery, and Enforcement Act of 1989, the dotcom, Enron, and WorldCom scandals with the Sarbanes Oxley Act of 2002, and the 2008 crash with the Dodd Frank Act of 2010. The Supreme Court then narrowed the prosecution tools in Arthur Andersen LLP v. United States in 2005 and Skilling v. United States in 2010. Detailed Findings What did the six debacles have in common? In each of the six debacles, a government decision widened the room for risk, and supervisors who could see the risk growing allowed fraud to grow with it. The sequence runs in four steps: a statute or rule grants permission, supervisors tolerate the resulting excess, insiders commit fraud that hides the losses, and Congress writes a repair statute after the public pays. Each permission carries a date. The Garn St Germain Depository Institutions Act of 1982, 96 Stat. 1469, opened thrift balance sheets to commercial lending. The Telecommunications Act of 1996, 110 Stat. 56, invited a telecommunications construction boom. The Gramm Leach Bliley Act of 1999, 113 Stat. 1338, and the Commodity Futures Modernization Act of 2000, 114 Stat. 2763, removed the barriers between commercial banks, investment banks, and insurers and placed over the counter derivatives outside federal and state regulation. Section 162(m) of the Internal Revenue Code, added in 1993, capped the deduction for executive salary at $1 million and pushed pay into stock options. The tolerance is equally specific. The Federal Home Loan Bank Board lowered the thrift net worth requirement from 5 percent to 3 percent between 1980 and 1982. The Securities and Exchange Commission adopted a voluntary capital supervision program for investment bank holding companies in 2004. The Federal Reserve held authority under the Home Ownership and Equity Protection Act of 1994 to set mortgage lending standards for all lenders and did not use it before the housing market peaked. The commission that investigated the 2008 crisis concluded that regulators possessed ample power and chose inaction. The repairs came after the losses. The cost of the six events ranges from the $124 billion that taxpayers absorbed in the thrift crisis to the roughly $5 trillion in market value that the technology stock collapse destroyed, with Enron shareholders alone losing about $74 billion. The rest of this report takes each event in turn, then compares what government action did before the losses, what it failed to do during them, and what the courts allowed afterward. How did government action cause and worsen the Savings and Loan Crisis? Government action caused the Savings and Loan Crisis in two stages: Congress and regulators widened what thrifts could do while shrinking the capital behind them, and then regulators let insolvent thrifts keep operating. Of the 3,234 federally insured thrifts in the system, 1,043 failed between 1986 and 1995, holding $519 billion in assets, and the cost to taxpayers reached about $124 billion by 1999. The first stage began with the Depository Institutions Deregulation and Monetary Control Act of 1980, 94 Stat. 132, which phased out interest rate ceilings on deposits and raised federal deposit insurance from $40,000 to $100,000 per account. Insured deposits meant depositors had no reason to examine the thrift that held them. The Garn St Germain Act of 1982 then allowed federally chartered thrifts to hold commercial real estate, commercial loans, and consumer credit. Residential mortgages fell from about 80 percent of thrift assets in 1980 to about 60 percent in 1986, and Congress added no matching examination staff. Between 1982 and 1985, thrift assets grew 56 percent, more than double the 24 percent growth rate at commercial banks. The capital side moved the other direction. The Federal Home Loan Bank Board cut the minimum net worth ratio from 5 percent to 3 percent, and regulatory accounting principles allowed thrifts to defer losses and count goodwill as capital. Under those rules, an investor could control a new institution with $2 million in capital and $1.3 billion in assets. Officials later called these practices accounting gimmicks. The second stage turned a funding problem into a solvency disaster. Insured losses grew while the Federal Savings and Loan Insurance Corporation held far too little in reserves: in 1983 it had about $6 billion against an estimated $25 billion needed to pay off the insured depositors of failed institutions. The Competitive Equality Banking Act of 1987, 101 Stat. 552, gave the insurer $10.8 billion through bonds and required forbearance, which allowed insolvent thrifts to remain open. An estimated $60 billion of the final cost traces to that forbearance. Failures peaked in 1988, with about 40 percent of them in Texas. Fraud sat on top of the policy failure. Estimates place fraud at 10 percent to 15 percent of the losses, or roughly $16 billion to $24 billion, with studies of individual failures finding fraud in 11 percent to 33 percent of them. Lincoln Savings and Loan, owned by Charles Keating through American Continental Corporation, was seized on April 14, 1989. More than 21,000 mostly elderly investors lost about $285 million in uninsured bonds sold in Lincoln branch offices, and the federal government absorbed close to $3 billion of Lincoln losses. Five United States senators met with regulators on Keating’s behalf while the examination of Lincoln was pending. Keating pleaded guilty to federal fraud charges in 1999. Justice Department figures for 1988 through 1992 record 2,603 convictions in major thrift fraud cases and 1,706 prison sentences. The Financial Institutions Reform, Recovery, and Enforcement Act of 1989, 103 Stat. 183, abolished the Federal Home Loan Bank Board and the Federal Savings and Loan Insurance Corporation, created the Office of Thrift Supervision and the Resolution Trust Corporation, raised capital standards, and removed goodwill from regulatory capital. The Resolution Trust Corporation disposed of 747 thrifts with more than $400 billion in assets before it closed on December 31, 1995. Section 951 of that statute, codified at 12 U.S.C. § 1833a, carries a civil penalty that the Justice Department revived in 2013 against a bank that sold defective mortgage securities. The largest losses to the insurer came from regulators delaying the closure of insolvent thrifts, and fraud explains a minority of the total. How did government action and fraud shape the dotcom crash of 2000 to 2002? Government action fed the dotcom bubble through tax, telecommunications, and monetary policy, and fraud by investment banks and issuers kept the bubble inflated until the Nasdaq Composite fell 78 percent from its March 10, 2000 peak of 5,048.62 to 1,114 on October 9, 2002. The decline erased about $5 trillion in market value. Three federal decisions supplied fuel. The Telecommunications Act of 1996 opened local telephone markets to competition and set off a fiber optic construction race that left capacity unused for years. The Taxpayer Relief Act of 1997, 111 Stat. 788, cut the top federal capital gains rate from 28 percent to 20 percent, which raised the after tax value of a successful stock sale. The Private Securities Litigation Reform Act of 1995, 109 Stat. 737, raised pleading standards for investor class actions and made it harder for investors to sue after a stock fell. The Federal Reserve cut rates three times in the fall of 1998 after a hedge fund collapse, then raised its target from 4.75 percent to 6.5 percent in six steps between June 1999 and May 2000, after the bubble was already inflated. Fraud entered through the initial public offering process. The ten investment firms that settled with the Securities and Exchange Commission and other regulators on April 28, 2003 in the Global Research Analyst Settlement agreed to pay about $1.4 billion: $487.5 million in penalties, $387.5 million in disgorgement, $432.5 million to fund independent research, and $80 million for investor education. Regulators alleged that between mid 1999 and mid 2001, investment banking pressure shaped analyst research, that firms paid for research without disclosing it, and that firms allocated shares of hot offerings to executives of companies whose banking business they wanted. That last practice is called spinning. Accounting problems also appeared at the peak. MicroStrategy restated its revenue on Mar

  7. 2d ago

    The 2026 Venture Fund Anti Money Laundering Landscape

    No federal law requires venture funds to establish a formal anti money laundering compliance program in 2026, because FinCEN delayed the Investment Adviser AML Rule until January 1, 2028. The final rule making that change was published at 91 FR 36 on January 2, 2026. An exemptive relief order issued on August 5, 2025 had already suspended every requirement of the rule in the interim. A venture fund adviser has no program, reporting, or recordkeeping duty under 31 CFR Part 1032 for the next seventeen months. The rule that was supposed to arrive in January 2026 was published on September 4, 2024 and added investment advisers to the definition of financial institution under the Bank Secrecy Act. It reached exempt reporting advisers, the category that includes any adviser managing only venture capital funds under Section 203(l) of the Investment Advisers Act of 1940. That was the part worth attention. The smallest and least examined adviser category in private capital was going to carry a five component program obligation under 31 CFR 1032.210(b): written policies and controls, independent testing, a designated compliance officer, ongoing employee training, and risk based ongoing customer due diligence. Then it was not. why does the delay not mean nothing changed Treasury put the highest illicit finance risk in the adviser sector on exempt reporting advisers in its February 2024 sectoral assessment, and named venture capital funds as a route foreign states use to reach early stage technology companies. A review of select Bank Secrecy Act reporting filed between January 2019 and June 2023 identified more than 20 investment advisers located in the United States advising private funds where the adviser had significant ties to Russian oligarch investors or to Russian linked illicit activity. That finding did not get withdrawn along with the compliance date. Sanctions law never depended on the Bank Secrecy Act in the first place. The sanctions regulations at 31 CFR Chapter V apply to every United States person on a strict liability basis. A fund that admits a limited partner owned fifty percent or more by a blocked person has dealt in blocked property whether or not it screened. The criminal provisions at 18 USC 1956 and 18 USC 1957 reach any person who transacts in criminally derived property, and willful blindness supports the knowledge element. A general partner who avoids asking where a large wire came from has not built a defense. He has built the government’s case. Offshore vehicles carry their own program duty. A fund organized in the Cayman Islands must appoint an Anti Money Laundering Compliance Officer, a Money Laundering Reporting Officer, and a Deputy Money Laundering Reporting Officer, with the reporting officer and the deputy being two different natural persons. Those appointments are notified to the local monetary authority and are untouched by anything Treasury did. Europe runs its own clock as well: Regulation (EU) 2024/1624 applies directly across all twenty seven member states from July 10, 2027. what actually forces a program today The binding constraint arrives by contract, not by regulation. In reviewing subscription documents and side letters for emerging manager venture funds over the past two years, I have watched the anti money laundering and sanctions representations expand well past anything a statute compelled. I have negotiated side letters where an institutional limited partner conditioned its commitment on the general partner maintaining a written anti money laundering policy, naming a compliance officer, and obtaining annual independent testing. Three of the four statutory pillars, imported into a private contract by a pension investor that did not care what the effective date of Part 1032 was. The enforcement mechanism is not an examination. It is a representation that stops being true, a default under the limited partnership agreement, and a diligence question at the next fund raise that the manager cannot answer well. Banks and administrators reach the same result from the other direction. A fund cannot operate without a deposit account, and the bank holding that account has its own customer due diligence duty that runs through to the investor base. Administrators run investor onboarding as a regulated service and will not accept a subscription without documented source of wealth, sanctions screening, and politically exposed person checks. what should a venture fund build now and leave alone Build what carries present legal consequence and will survive any rewrite. Screening of every limited partner, every beneficial owner at the twenty five percent threshold, and every portfolio company against the sanctions lists, because strict liability offers no deferral. Documented source of wealth for individual subscribers above a threshold the fund sets for itself. A short written policy. A named compliance officer located in the United States. Retention of subscription and screening records for five years. Leave the rest for now. On April 7, 2026 the agency issued a proposal rewriting Bank Secrecy Act program requirements for every category of financial institution, replacing technical compliance with an effectiveness standard and separating program design from program operation. Comments closed June 9, 2026, and the proposal contemplates a twelve month implementation period after any final rule. Writing detailed suspicious activity reporting procedures, Section 312 correspondent account protocols, and travel rule mechanics against 2024 rule text that the promulgating agency has publicly committed to revising spends the same compliance dollar twice. There is also reason to think the effectiveness standard helps venture managers specifically. A fund with fifty limited partners, most of them repeat institutions already documented elsewhere, has a low risk investor base by any measure. A rule that directs attention and resources toward higher risk customers and away from lower risk ones lets that fund write a short program and defend it, instead of buying transaction monitoring designed for a retail bank. what is still unknown Two things. Whether Treasury narrows the covered adviser definition to drop venture advisers before the 2028 date, which commenters asked for and the agency declined to address while promising to tailor the rule later. And whether the customer identification program proposal that the agency issued jointly with the Securities and Exchange Commission on May 21, 2024 comes back in the same shape. It was dropped from the Securities and Exchange Commission’s 2026 rulemaking agenda after sitting at the final rule stage. The second question is the one with money attached, and it is the subject of the next piece: when the customer identification proposal is reissued, does a limited partner count as a customer whose identity the manager must independently verify, or does reliance on a fund administrator satisfy the rule. That single answer decides whether investor onboarding becomes a general partner expense line or stays with the service provider, and it will move a venture fund’s compliance budget more than the program rule ever will. Read my full analysis here: This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit theinnovationattorney.substack.com/subscribe

  8. 2d ago

    Oklo Natural Nuclear Reactors

    Executive Summary The Oklo natural nuclear reactors in Gabon, discovered in 1972, sustained uranium fission chain reactions about 2 billion years ago and remain the only confirmed natural reactors on Earth. Sixteen reactor zones across the Oklo, Okélobondo, and Bangombé uranium deposits reached criticality because natural uranium then held about 3.5 percent uranium 235 and groundwater slowed neutrons inside ore that contained very little neutron absorbing material. Boiling water shut each zone down and returning water restarted it, a cycle that xenon isotopes trapped in the mineral place at roughly 30 minutes of fission followed by about 2.5 hours of rest. Many fission products and actinides stayed within or near the ore for the full 2 billion years, which gives repository designers a field record that no laboratory test can replicate. See Sources 1, 3, and 20. The subject carries commercial weight in 2026. A Department of Energy authorized isotope reactor built by a company named for the Gabon site reached criticality in Lockhart, Texas, on August 5, 2026, and the Finnish radiation regulator found no obstacle to an operating license for the Onkalo spent fuel repository on August 10, 2026. Onkalo is described as the first commercial final disposal facility for spent fuel, so its retention evidence must come from analogues, and Oklo supplies a record measured in billions of years. See Sources 7, 16, and 18. Two legal instruments set the stakes. Section 63.311 of Title 10 of the Code of Federal Regulations limits dose to the reasonably maximally exposed individual to 15 millirem per year for the first 10,000 years and 100 millirem per year through one million years, and 42 U.S.C. § 2210 extends Price Anderson Act liability protection under the 2024 amendments through December 31, 2065. See Sources 9 and 15. Detailed Findings The Oklo record rests on four lines of evidence: isotope depletion in the ore, fission product signatures, xenon trapped in minerals, and neutronics simulations of the reactor zones. Each line was developed by separate laboratories over five decades, and they converge on the same operating history. See Sources 1, 2, and 3. What are the Oklo natural nuclear reactors and where are they located? The Oklo natural nuclear reactors are sixteen zones of self sustaining uranium fission in the Franceville basin of Gabon, spread across the Oklo, Okélobondo, and Bangombé deposits. See Source 20. One 2011 neutronics study counts approximately fifteen zones at the Oklo and Okélobondo mines. See Source 3. Each zone is a lens of uranium ore a few tens of centimeters thick, set in sandstone and clay. The best studied zone, labeled RZ9, measures about 12 meters long and 7 meters wide. See Source 3. That study places operation near 1.95 billion years ago, and other compilations give a window between 1.7 and 2 billion years ago. See Sources 3, 19, and 21. How was the Oklo natural reactor discovered in 1972? Routine mass spectrometry on uranium hexafluoride samples at the Tricastin enrichment site in Pierrelatte, France, revealed uranium 235 depletion in Oklo ore in May 1972, and the French Atomic Energy Commission announced the finding on September 25, 1972. See Source 21. Natural uranium on Earth carries 0.72 percent uranium 235. One early sample read 0.717 percent against the 0.720 percent standard, a gap small enough that investigators first looked for an artificial cause until physicist Francis Perrin and colleagues established that the process was natural. See Source 19. Some ore samples later measured as low as 0.44 percent, about 40 percent below normal. See Source 21. The missing uranium 235 appeared as fission products. Natural neodymium contains 27 percent neodymium 142, while the neodymium extracted from Oklo ore contains less than 6 percent, and ruthenium 99 reaches 27 to 30 percent against 12.7 percent in nature because technetium 99 decayed into it. See Source 21. Why did natural fission occur at Oklo about 2 billion years ago? Natural fission occurred at Oklo because five conditions coincided: enriched uranium, concentrated ore, water, adequate size, and low neutron absorption. A reactor needs all five at once, and no other known deposit has met them. See Sources 2 and 3. Enrichment came from radioactive decay. Uranium 235 decays with a half life of 704 million years and uranium 238 with a half life of 4.47 billion years, so the uranium 235 share of natural uranium stood near 3.5 percent about 1.95 billion years ago, compared with 0.72 percent today. See Source 3. Light water reactors run on fuel enriched to a similar range, which is why ordinary groundwater could serve as moderator. Concentration required ore with at least about 20 percent uranium by weight in dehydrated form. See Source 3. Atmospheric oxygen had risen enough by then to dissolve uranium as the uranyl ion, carry it in groundwater, and deposit it where reducing conditions trapped it. See Sources 2 and 21. Water slowed fast neutrons to thermal energies, where uranium 235 fission is most probable. See Source 20. Simulated critical radii ranged from 50 to 160 centimeters depending on geometry, and the best studied lens is several times wider than that range. See Source 3. The ore held little of the boron and rare earth material that absorbs neutrons. See Source 2. In 1956 the physicist Paul Kuroda calculated that these conditions could produce a chain reaction in a uranium deposit, sixteen years before the discovery in France. See Source 4. At 0.72 percent, natural uranium cannot sustain a chain reaction in ordinary water, so a natural reactor cannot start today without heavy water or graphite as moderator. See Source 21. How did the Oklo reactors regulate themselves for 150,000 years? Water governed each Oklo reactor zone as both moderator and thermostat. Fission heated the zone, the water boiled away, the chain reaction stopped, and cooling let groundwater return to restart it. See Source 1. Xenon isotopes trapped in aluminum phosphate grains record the rhythm: roughly 30 minutes of criticality followed by about 2.5 hours of cooling, a three hour cycle. See Sources 1 and 21. Average thermal power stayed below 100 kilowatts. About five tons of uranium 235 fissioned over the life of the zones, releasing roughly 420 petajoules, an energy equivalent to about 100 megatons of TNT. See Source 21. Estimates of total duration range from nearly 100,000 years through about 150,000 years to a few hundred thousand years, depending on the zone and the method. See Sources 3, 20, and 21. On a nuclear submarine the reactor operator learns to respect xenon 135, a fission product that absorbs neutrons so strongly that a reactor shut down for several hours can refuse to restart until the xenon decays. The same element, preserved in mineral, recorded the operating history of a reactor that ran two billion years before anyone built a control panel. The loss of moderator through boiling that stopped each Oklo cycle is the negative void feedback that engineers design into light water reactors today. What do the Oklo reactors show about fission product and plutonium retention? Oklo shows that many fission products and actinides stayed within or near the reactor zones for about 2 billion years, held in uraninite and aluminum phosphate minerals with clay around the zones. Phosphate rich minerals such as apatite retained actinides and fission products effectively, and uranium showed little or no migration. See Sources 8 and 20. Retention varied by element. Rare earth fission products and uranium largely stayed, and some volatile and soluble elements dispersed, so analogue studies report retention element by element. See Sources 2 and 8. Plutonium 239, produced when uranium 238 captured neutrons, has a half life of 24,100 years, so none remains. Its decay product, uranium 235, stayed in the ore, and that excess marks where plutonium once sat. See Source 2. Neodymium belongs to the fission products, and the Oklo neodymium isotope mix carries the fission signature described above. See Source 21. The analogy has limits. Oklo was a natural ore body at modest temperatures. An engineered repository such as Onkalo adds iron copper canisters, a bentonite buffer, and clay backfill, and spent fuel gives off decay heat for centuries. See Source 18. The Oklo record validates chemical retention in a geological setting and does not test canister corrosion or bentonite performance. How did Oklo test whether physical constants change over time? Oklo isotope ratios let physicists bound any change in the fine structure constant over two billion years, and analyses published through 2014 conclude that nuclear reactions then were much the same as they are today. See Sources 5 and 21. That result applies a reactor built by geology to a question in fundamental physics, and it supports the use of present day nuclear data to model the zones. How does Oklo connect to current reactor developers and waste policy? Oklo Inc., founded in 2013, takes its name from the Gabon site. See Source 21. The Nuclear Regulatory Commission denied the Aurora combined license application of the company on January 6, 2022, without prejudice, under docket 05200049. See Source 6. The Department of Energy selected the company for its Reactor Pilot Program in May 2025, and the Groves Isotope Test Reactor reached criticality at 9:19 p.m. Eastern time on August 5, 2026. See Sources 7, 16, and 21. Groves is a pool type, water cooled, nonpressurized reactor of 15 megawatts thermal using metal fuel, the fifth Department of Energy authorized advanced reactor to reach criticality in summer 2026 and the first on private land. See Sources 7 and 16. The Aurora powerhouse broke ground at Idaho National Laboratory in September 2025, and the Idaho Radiochemistry Laboratory received an NRC materials license. See Source 17. Finland supplies the disposal counterpart. On August 10, 2026, the Finnish Radiation and N

See All (235)

About

The Innovation Attorney is a publication about the intersection of venture capital, technology, public policy and law. theinnovationattorney.substack.com

Information