NEW YORK, November 22, 2026: Anthropic’s record-breaking initial public offering has exposed tensions between major technology companies and Wall Street banks, with unusually strict confidentiality requirements, limited deal teams and last-minute document changes complicating the listing process.
The artificial intelligence company went public on Thursday in an offering internally known as Project Lantern, surpassing SpaceX’s June IPO and generating more than $500 million in fees for participating banks.
Behind the successful listing, however, several bankers and advisers expressed frustration over how the transaction was managed. People familiar with the process told Reuters that Anthropic imposed restrictions rarely encountered in major public offerings, including limits on staffing and requirements affecting bankers’ future work with competing technology companies.
Some of the 22 underwriting banks were instructed to assign no more than four bankers to the transaction, with Anthropic requesting senior personnel for those roles.
People involved in the deal suggested that concerns about confidential information reaching the media influenced the restrictions. Banks nevertheless found ways to manage the workload despite the staffing limits.
Anthropic also asked certain senior banking officials to sign agreements preventing core transaction team members from working with competitors for six months after the listing.
The restrictions covered companies including OpenAI, Cursor, DeepSeek and Hugging Face.
Such arrangements are uncommon in IPO transactions, where banks generally manage potential conflicts through established confidentiality and compliance procedures.
The requirements presented difficulties for bankers whose responsibilities included maintaining relationships with other artificial intelligence companies considering public offerings or additional fundraising.
Information security was another significant feature of the transaction.
Rather than distributing conventional PDF documents, Anthropic used NetRoadshow, a platform that records document access and tracks which pages individual users view.
The materials also carried watermarks identifying bankers, while restrictions on saving copies complicated document reviews.
The measures followed the publication of details from Anthropic’s confidential IPO filing nearly four weeks before the document became publicly available.
The company’s preparations for its November 10 marketing launch also encountered difficulties.
A law firm working on the offering initially circulated a document containing incorrect information about the proposed share price range and share allocations relating to Canada.
Recipients were subsequently instructed to disregard the document.
A corrected securities filing was distributed through NetRoadshow at approximately 2 a.m. New York time, leaving participating banks only several hours to review and approve it before submission to the US Securities and Exchange Commission.
Anthropic’s approach drew comparisons with SpaceX’s June listing, when Elon Musk’s company exercised extensive control over marketing arrangements, investor selection and underwriting procedures.
Both companies also negotiated arrangements under which underwriters received no fees on additional shares sold through the overallotment option, according to University of Florida professor emeritus Jay Ritter.
The strong demand for shares in both offerings gave the companies considerable negotiating leverage over financial institutions seeking to participate.
Despite the operational difficulties, both IPOs attracted substantial investor interest and were heavily oversubscribed.
Their subsequent stock performance, however, showed different trajectories.
SpaceX shares reached $201.80 on June 16 before falling below their $135 offering price in mid-July.
Anthropic shares gained 17 percent on their first trading day but closed 8.5 percent lower on Friday amid broader weakness in artificial intelligence-related investments.
Anthropic, Goldman Sachs and Morgan Stanley declined to comment on the reported concerns, while SpaceX and OpenAI did not respond to requests for comment.

