The AI lab's June 1 confidential S 1 is thin on details, but the three part governance architecture it claims to operate under — Delaware public benefit corporation status, the Long Term Benefit Trust, and the Responsible Scaling Policy — is about to be stress tested by the…
On June 1, 2026, Anthropic announced that it had confidentially submitted a draft S-1 registration statement to the U.S. Securities and Exchange Commission. The notice, issued under Securities Act Rule 135, is the bare-minimum communication a company is allowed to make around an impending IPO: it confirms the filing exists, identifies the issuer, and says nothing about share count, price range, exchange, or timing. The document is not yet public, and the company is not yet offering securities. What it does do, for the first time, is force Anthropic to start explaining its governance to readers who are no longer just customers, employees, and limited partners — namely, the public.
That is the part worth watching. Anthropic has spent five years building a self-described three-part governance architecture: its Delaware Public Benefit Corporation (PBC) charter, which commits the company to a specific public benefit purpose; the Long-Term Benefit Trust, a separate purpose trust that holds a special class of stock and elects a slice of the board; and the Responsible Scaling Policy (RSP), an internal framework for when to slow or stop training and deployment as model capabilities grow. The confidential S-1 does not change any of those documents on its own. But it does begin the process by which all three of them will be read, parsed, and litigated against by a much wider set of actors than Anthropic has faced before.
Anthropic is organized as a Delaware Public Benefit Corporation. Its certificate of incorporation states a specific public benefit purpose: "responsibly develop and maintain advanced AI for the long-term benefit of humanity," as described by Anthropic's outside counsel at Wilson Sonsini Goodrich & Rosati in a 2023 Harvard Law Forum on Corporate Governance post. Delaware PBC law requires directors to balance the public benefit purpose against shareholder interests — they do not have to put shareholders first, but they also do not have to subordinate shareholders to the mission. That distinction matters once a company has public shareholders with standing to sue.
Sitting alongside the PBC charter is the Long-Term Benefit Trust, a Delaware common-law purpose trust with five Voting Trustees distinct from the company's board. The trust holds a special Class T Common Stock and, under the 2023 design, elects a gradually increasing number of Anthropic's directors — initially 1 of 5, then 2, and eventually a majority of 3 on time- and fundraising-based milestones. The trust also receives advance notice of certain board actions. Anthropic has framed the trust as the structural answer to a question the rest of the AI industry has struggled with: how do you keep a company honest about a safety mission once the cap table fills up with investors who want a return?
That framing has limits. As the 2023 Harvard Law Forum post notes, the seat allocation and trustee-selection mechanics have evolved over subsequent fundraising rounds; the 1-of-5 starting point and the 3-of-5 endpoint are the on-record design from 2023, and any post-Series-D amendments would have to be reconciled against the public S-1 when it lands. More fundamentally, the trust has no public track record of overriding management. It has never had to fire a CEO, block a release, or refuse a financing. The IPO will be the first event that meaningfully tests whether the trust's structural rights translate into actual friction inside the company.
The third leg of Anthropic's governance claim is the Responsible Scaling Policy — an internal, versioned framework that ties further training and deployment to capability thresholds and safety evaluations. The RSP is not a contract; it is a policy Anthropic has committed to follow, and the company has revised it under commercial pressure in the past. What changes under public-company disclosure is that any future RSP revision becomes a material event that may need to be explained in filings, in earnings communications, and to shareholders who can ask, in writing, why a commitment was loosened. That is a different accountability regime than the one that produced the current RSP.
Going public does not, by itself, change a PBC's statutory duties. Anthropic's directors will still be required to balance the public benefit purpose against shareholder interests after the IPO, just as they were before it. What changes is who is on the other side of that balance. Pre-IPO, Anthropic's shareholders were a small group of mission-aligned investors and employees. Post-IPO, the cap table will include public holders with standing to bring fiduciary suits, securities-fraud actions, and books-and-records demands. The prospectus will have to disclose related-party transactions, conflicts policies, and board composition in a way the company has never had to do before. The S-1 process is, in effect, the first independent audit of the governance story Anthropic has been telling.
The academic framing for what Anthropic's structure is built to address comes from a Harvard Law Review article, "Amoral Drift in AI Corporate Governance" (Vol. 138). The article borrows the term "amoral drift" — originally from economists Hart and Zingales — to describe the slow erosion of a company's stated mission under sustained market pressure, and uses the 2023 OpenAI board saga as a cautionary case of what happens when a stakeholder mandate fails in practice. The point is not that Anthropic is currently drifting; it is that the pressure to drift is structural, and the governance mechanisms a company adopts are only as durable as their ability to resist that pressure under cost.
When the public S-1 lands, the concrete signals worth looking for are:
These are not predictions. They are the items a public-benefit-corporation S-1 is structurally required to surface, and they are the items the Long-Term Benefit Trust's design is supposed to make legible.
Anthropic is not the only frontier-AI company approaching the public markets. As Wired reported on June 1, 2026, SpaceX (the parent of xAI) filed confidentially in April 2026 and is targeting a June 12 debut at a $1.75T valuation; OpenAI is rumored to be weighing an offering as soon as September 2026. Wired also flagged Anthropic's pre-IPO valuation context — a $965B figure, a $65B Series H raise, and a $47B annualized revenue number that originate from Anthropic's own Series H announcement — and described the PBC and Long-Term Benefit Trust structure as something that "could lead to both delays and a knockdown in valuation." Those are the company's self-reported numbers, not independent market data, and the filing will be the first place they are subjected to prospectus-level verification.
The financial profile is unusually concentrated. SAN reported in June 2026 that Anthropic's near-100% R&D-to-revenue burn rate compares with roughly 15% for Alphabet, and framed the IPO as forcing a choice between shareholder primacy and a dual-class structure of the kind Google adopted at its own listing. That framing is the wire-press read on the same question this piece is asking: whether the governance architecture survives the cap-table change.
Anthropic is filing while in active friction with the U.S. government. As SAN and Wired both reported in June 2026, Defense Secretary Pete Hegseth sanctioned the company under supply-chain laws, citing its refusal to remove safeguards from Claude for military use; Anthropic sought to challenge the blacklisting, and a judge has denied its appeal. The status of that litigation is likely to be a disclosure item in the public S-1, and the company's stated rationale for refusing the requested changes will be tested against the prospectus's standard for accuracy.
This is not a story about whether Anthropic is a good company, whether the IPO will succeed, or whether public-benefit-corporation structures are a good idea. It is a story about whether a specific set of governance documents — drafted when Anthropic was private, small, and selectively funded — will hold up under the legal and market scrutiny that comes with a public cap table. The confidential S-1 does not answer that question. The public S-1, when it comes, will start to.