On October 14 the Federal Trade Commission sent OpenAI and Anthropic demands of more than 50 pages each about safety incidents, safeguards and marketing, Axios reported. The same day eight states sued OpenAI in their own courts, and New York City's comptroller, who oversees the city's pension funds, asked Anthropic to pause its planned public offering. Anthropic, meanwhile, told investors at a closed meeting, according to attendees, that its unaudited third-quarter revenue was $16.8 billion; a day later Reuters reported investors discussing a valuation near $1.5 trillion rather than $2 trillion.
None of these questions is new. The FTC questioned OpenAI in 2023, Florida sued it in June, and every company in registration answers the SEC privately before its prospectus is public. OpenAI says it will not list this year, so for it these are consumer-protection questions. For Anthropic, which is in registration and limited in what it may say publicly, they are also questions its prospectus must answer, asked in public by people other than the SEC, about numbers and claims it largely put out itself. Modal raised at $15.75 billion on a run rate it reports itself, and its announcement does not say how that rate is counted. The questions a prospectus answers can arrive before it. Write your answers while your record is still short.
Modal announced a $750 million Series D led by Accel on October 14, at a $15.75 billion post-money valuation according to Pillsbury, its counsel, with Kleiner Perkins joining Menlo, General Catalyst, Redpoint, Lux and Bain Capital Ventures. Modal sells serverless compute for inference, fine-tuning, batch jobs and sandboxes for agent-written code, and says its annualized run rate has passed $1 billion, with sandboxes alone near $300 million. The terms match what TechCrunch reported on September 28 as a round Modal was closing in on, and the valuation is more than three times May's $4.65 billion. The headline is a valuation that tripled in five months. The more useful read is that nothing in the announcement says how the run rate is counted, and the round priced anyway. That question was put to the largest labs in public this week; expect Modal to meet it at its next round or listing.
The Financial Stability Board published its final report on sound practices for responsible AI adoption on October 15: twelve non-binding practices for financial institutions covering board oversight, governance, data, explainability, human oversight, cyber risk and dependence on third-party AI providers. They are guidance, not rules. They are also the likeliest route by which this week's questions reach a startup, through the risk questionnaires banks send their vendors.
The FTC sent OpenAI and Anthropic compulsory demands on October 14, Axios reported, of more than 50 pages each covering safety and security incidents, safeguards, marketing materials and privacy policies, as part of a consumer-protection inquiry; an FTC spokesperson confirmed the subpoenas to CBS. The agency plans to take testimony from top executives, including some at METR, an outside model evaluator. The step was expected: Semafor reported the probe on September 30, and the FTC first questioned OpenAI in 2023. It is, though, the first consumer-protection demand reported against Anthropic, arriving while it is in registration.
On October 14 the attorneys general of Arizona, Arkansas, Idaho, Iowa, Kansas, Montana, Nebraska and West Virginia filed separate suits against OpenAI in their own state courts, under consumer-fraud and public-nuisance theories centred on minors; the complaints allege ChatGPT reinforced delusions and encouraged self-harm. These are allegations, not findings, and OpenAI points to its age-specific protections. Florida sued alone in June. The states carry the burden of proof, but OpenAI now defends its safety claims in eight separate proceedings, the kind of pending litigation any company must disclose if it lists.
In an October 14 letter, Comptroller Mark Levine, who oversees New York City pension funds of more than $325 billion, urged Anthropic to “pause its planned initial public offering” until it addresses safety, governance and development-pace concerns, and to disclose “with specificity” how it audits and controls its models. His office says the funds have about $7 million of indirect exposure (about 0.002% of the funds) and would become holders through index inclusion. Anthropic filed a confidential draft registration in June; IFR reports a public filing planned for October 26. It is a question from the city workers' funds that may one day own the stock, asked before the document they could read.
At a closed investor meeting on October 14, Anthropic said its unaudited third-quarter revenue was $16.8 billion, according to the FT, Bloomberg and Reuters, which cite attendees; the company has not published the figure and declined to comment to Bloomberg. That is about 45% above the roughly $11.6 billion Reuters reports for the second quarter, after revenue more than doubled the quarter before. On October 15 Reuters, citing four people, reported investors discussing a valuation near $1.5 trillion against the roughly $2 trillion previously expected, citing disappointment at slowing growth. Investors discussing a number is not a price; the disclosure was Anthropic's own answer, and the grade comes when the offering prices.
Paul Kedrosky told Axios on October 13 that annualized revenue metrics at the largest AI companies are “inherently arbitrary and capricious.” The prompt was an October 8 FT report that OpenAI told investors its annualized revenue was approaching $50 billion, against roughly $70 billion in earlier reports; the reports attribute the gap to investors restating OpenAI on Anthropic's basis, which counts sales made through cloud partners. It is one commentator's view, not a finding, but it names the question Anthropic's prospectus will have to settle in audited terms.
Axios reported on October 14 that a16z laid off most of the 34-person talent team it has offered portfolio founders as a central recruiting service, moving the remaining members into individual investing groups; the firm declined to comment and has not said it is ending recruiting help. A platform team is a claim in an investor's pitch, and like a self-reported run rate it is worth asking how it is counted: which people, doing what.
At TechCrunch Disrupt, Benchmark's Eric Vishria said “you are going to get conflicts, and that's just the reality,” and Peter Fenton said the rule now is transparency, so that there is no surprise after the fact; TechCrunch reported the exchange on October 15 as AI eroding the venture pact against backing rivals. TechCrunch's example is Factory and Cognition, rivals that Khosla Ventures backs both of. Benchmark says exclusivity still matters at the earliest stage. When an investor may back your competitor, the questions you ask before granting information rights carry more weight.
Investors funded an estimated 8,625 first-time US financings through the third quarter, above the 8,507 of all 2021, with $34.3 billion invested, according to PitchBook data. PitchBook reports smaller funds now courting founders who have not yet founded anything. Money is arriving earlier in a company's life, which means first diligence happens more often on a thin record.
Thoma Bravo-owned Proofpoint signed a definitive agreement on October 15 to buy Varonis Systems for $49.75 a share in cash, about $6.3 billion, which Varonis's filing puts at 8.4 times trailing revenue on an enterprise-value basis; closing is expected in the first half of 2027. The companies say the combination creates the industry's largest data and AI security organization. Varonis is a public company with audited numbers, a record a cash buyer can price without taking the seller's definitions on trust.
Workday said on October 13 that it will acquire C1.ai, which grants and revokes access for employees and AI agents across email, CRM, code repositories, finance tools and cloud accounts; terms were not disclosed, and closing is expected by January 31, 2027. Workday is buying a way to answer a question buyers are likely to ask before an agent pilot starts: what can it reach, and who can revoke it.
CoreWeave's 2032 notes yielded about 14.8% and briefly touched 15% on October 13 and 14, and its shares fell about 16% on October 14, before the bonds partly recovered. Bloomberg's tally of filings found the three co-founders have sold about $3.3 billion of stock since the March 2025 listing, under pre-arranged plans, and co-founder Brannin McBee told the FT the company wants investment-grade ratings to reach insurers' capital. A prospectus does not end the questions; after a listing they are priced every trading day, and CoreWeave's bondholders are pricing whether debt-funded capacity can earn the rating it wants.
The IMF's Global Financial Stability Report, released October 13, names a correction in AI-linked assets as one of three rising risks, warning that faltering AI investment and profitability could trigger it. Its review of 14 transactions since 2023 counts about $300 billion of equity and $650 billion of debt invested in exchange for $1.4 trillion of future revenue commitments. The IMF frames it as a scenario, not a forecast; its question is the counting one again: when does a commitment become revenue.
This week looked like two AI labs in trouble. I read it as a timing story: the questions a prospectus exists to answer were asked in public before anyone had published one.
Half of my #032 ↗ wager was never a live test: the agent protocol draft I called due this month had been out since October 9, three days before I published, and it requires a customer's grant before a guest agent makes changes. The OPT half stays open. This week the FTC asked both labs about safety incidents, eight states sued OpenAI, and New York City's comptroller, who oversees the city's pension funds, asked Anthropic to wait. Anthropic volunteered one answer, a $16.8 billion quarter no filing has yet defined, and within a day investors were reportedly discussing $1.5 trillion rather than $2 trillion.
None of these questions is new. What is new is that they came in public, from people other than the SEC, while Anthropic is in registration. Modal raised at $15.75 billion on a run rate it reports itself, and nothing in its announcement says how it is counted.
Here is what would change my mind. If Anthropic's public registration statement reports its second- and third-quarter revenue within 5% of what it told investors, about $11.6 billion and $16.8 billion, its private numbers held up and definitions matter less than I am arguing. If nothing is public by December 31, the wager is void.
Write down how you count revenue, what has gone wrong and who can stop a launch while the record is short, and show investors the definition before the number. A definition written before anyone asks is a standard; one written after is a defense.