MAIN ARTICLE
The issuer takes the commanding seat
For decades, major investment banks have helped companies navigate public listings by assembling sales teams, testing investor appetite, examining disclosures and distributing shares. Anthropic’s deal inverted some of that customary relationship. People familiar with the transaction told Reuters that the company asked some of its 22 underwriters to assign no more than four bankers, preferably senior personnel, to the project. The precise arrangements were not publicly detailed by Anthropic, and accounts of their motivation come from people close to the process.
The transaction was known internally as Project Lantern. Its magnitude gave Anthropic considerable leverage: banks stood to earn substantial fees and gain an important relationship in a sector attracting intense capital. Professor emeritus Jay Ritter of the University of Florida told Reuters that such issuers had the market power to demand departures from standard underwriting practice. He also cautioned that only a very small number of exceptionally desirable companies might be able to repeat these terms.
Confidentiality becomes a bargaining condition
Reuters reported that certain senior banking officers were asked to sign six-month restrictions affecting work for competitors including OpenAI, Cursor, DeepSeek and Hugging Face. Confidentiality obligations and conflict controls are familiar in finance, but limits on individual bankers’ ability to work with multiple rival companies were described by people involved as atypically broad.
A banker who declined to sign said finding eligible colleagues was difficult because many had relationships across the fast-growing AI industry.
The company also distributed sensitive materials to many banks through NetRoadshow, a platform that can record document access and display personal watermarks. Those controls appeared intended to deter leaks after Reuters obtained details of an early draft filing weeks before public release. But at least one person involved said the inability to save ordinary copies made reviewing materials harder. Protecting sensitive information is legitimate; the challenge is ensuring protection does not interfere with careful verification by professional advisers.
A deadline measured in hours
The pressure became especially visible ahead of a planned November 10 amendment to Anthropic’s registration statement. According to Reuters, advisers circulated a Canadian offering document containing an incorrect price range and share information and then instructed recipients to disregard it. Around 2 a.m. New York time, the corrected S-1A appeared on NetRoadshow. Some banks were given only several hours to review and approve the draft before submission to the U.S. Securities and Exchange Commission.
Nothing in these reported accounts, by itself, establishes a violation of securities rules. Nevertheless, compressed review windows matter because a prospectus informs investors about risks, business performance and ownership. Underwriters must conduct appropriate due diligence even when an issuer insists on speed. The episode raises a governance question extending beyond one company: can a deal be both meticulously guarded and sufficiently open to challenge from those responsible for checking it?
A triumphant listing meets a skeptical market
The listing brought Anthropic to Nasdaq under the ticker ANT. Reuters reported an offer price of $83 a share and a first-day closing price of $97.17, roughly 17% higher. The shares subsequently slipped 8.5% on November 20 as wider technology markets weakened. The contrast is revealing: oversubscribed order books can demonstrate demand at issuance, but they cannot guarantee future prices or settle debates about the long-term economics of AI.
The pattern also echoes SpaceX’s June flotation, when people close to that process described firm control over marketing, allocations and fees. Reuters reported that both issuers negotiated no underwriting spread on shares sold through the additional-allotment, or greenshoe, option. Anthropic, Goldman Sachs and Morgan Stanley declined Reuters’ requests for comment; SpaceX and OpenAI did not respond. Their silence leaves the reported complaints as attributed accounts rather than a fully contested public record.
















