# Anthropic Long-Term Benefit Trust

> Source: https://aiwiki.ai/wiki/anthropic_long_term_benefit_trust
> Updated: 2026-07-27
> Categories: AI Companies, AI Ethics, AI Safety, Anthropic
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution to "AI Wiki (aiwiki.ai)".

The **Anthropic Long-Term Benefit Trust** (LTBT) is a Delaware purpose trust that holds a special class of [Anthropic](/wiki/anthropic) stock and uses it to elect a portion of the company's board of directors. Anthropic announced it on 2023-09-19 as an attempt to give people with no financial stake in the company a lasting say in who governs it [1]. The Trust's trustees hold no equity in Anthropic, do not share in its profits, and are compensated only for their time and service [12]. Its board-election power was designed to grow over time, and in April 2026 Anthropic said that directors appointed by the Trust had become a majority of its board [3]. Whether the arrangement actually constrains Anthropic when money and mission conflict is unresolved: the governing Trust Agreement has never been published, the thresholds at which stockholders can amend the Trust's powers are not public, and no public episode has turned on the Trust exercising its authority against the wishes of management or investors.

## What the Trust is

Anthropic is incorporated in Delaware as a [public benefit corporation](/wiki/ai_governance), with a stated public benefit purpose of "the responsible development and maintenance of advanced AI for the long-term benefit of humanity" [2]. The PBC form and the Trust are two separate mechanisms and are often conflated. The PBC form obliges directors to balance stockholder returns against the stated public purpose and gives them legal cover for doing so; it does not change who chooses the directors. The Trust does exactly that. It is a layer on top of the corporate form that changes the composition of the body making the balancing judgment.

The legal vehicle is a Delaware purpose trust with five Voting Trustees plus an Administrative Trustee, described in a 2023 analysis co-authored by Yale Law School's John Morley with David Berger and Amy Simmerman of Wilson Sonsini Goodrich and Rosati, the firm that helped design it [4]. The Administrative Trustee is a trust institution sited in Delaware holding a narrow set of administrative powers whose purpose is to secure Delaware jurisdiction and choice of law; when Anthropic and the authors say "Trustees" they mean the Voting Trustees [4]. Anthropic described the body as "an independent body of five financially disinterested members" whose remit is deliberately narrow: the Trust is meant to concern itself primarily with long-range issues and extreme outcomes rather than ordinary commercial decisions, on the reasoning that public benefit and stockholder returns usually point the same way in day-to-day operations [1].

New trustees are chosen by the sitting trustees, in consultation with Anthropic, and trustees serve annual terms subject to reappointment by their peers [1][4][5].

## Class T stock and the board phase-in

The Trust's power runs through a class of shares created for it, Class T common stock, added by charter amendment at the close of Anthropic's Series C [1][4]. The Class T shares carry board-election rights but, as Morley, Berger and Simmerman put it, "the financial claim represented by the Class T shares is very small" [4]. The Trust is a governance holder, not an economic one.

The election rights were structured to escalate on a mix of time-based and funding-based milestones: initially one director out of five, then two, then three, which would have been a majority of a five-member board [1][4]. Anthropic put an outer bound on the schedule in the same announcement: "in any event, the Trust will elect a majority of the board within 4 years" [1]. TIME reported in 2024 that the milestones pointed to two of five seats in July 2024 and three of five in November 2024, and that Anthropic had already passed the associated fundraising thresholds [6].

What complicated that schedule is that the board itself grew. By April 2026 Anthropic's board had seven members, so three Trust-appointed directors were no longer a majority. Anthropic stated the majority threshold was crossed only with the appointment of Novartis chief executive Vas Narasimhan on 2026-04-14: "With Narasimhan's appointment, Trust-appointed directors now make up a majority of the Board" [3].

That statement does not reconcile with the public record on its face. Anthropic has named only three directors as Trust appointments: Jay Kreps [25], Reed Hastings [8] and Narasimhan [3]. A majority of a seven-member board is four. Either one of the remaining directors is also a Trust appointee whom Anthropic has never identified as one, or the claim rests on a seat count the company has not published. R&D World made the same observation at the time, noting that the majority claim "on a seven-seat board implies a fourth appointment, though Anthropic has not specified which of the remaining directors, Dario Amodei, Daniela Amodei, Yasmin Razavi or former White House official Chris Liddell, falls on which side of the line" [7].

| Director | Joined | Appointed by |
|---|---|---|
| Dario Amodei | Co-founder | Stockholders |
| Daniela Amodei | Co-founder | Stockholders |
| Yasmin Razavi (Spark Capital) | Investor director | Stockholders |
| Jay Kreps (Confluent) | 2024-05-29 | Long-Term Benefit Trust [25] |
| Reed Hastings (Netflix) | 2025-05-28 | Long-Term Benefit Trust [8] |
| Chris Liddell (former Microsoft CFO) | 2026-02-13 | Not publicly specified [9] |
| Vas Narasimhan (Novartis) | 2026-04-14 | Long-Term Benefit Trust [3] |

As of 2026-07-27 Anthropic's company page lists six directors: Dario Amodei, Daniela Amodei, Yasmin Razavi, Reed Hastings, Chris Liddell and Vas Narasimhan [2]. Jay Kreps, who was on the board in April 2026, is no longer listed. Anthropic has published no announcement of his departure, and of the six directors now listed only two, Hastings and Narasimhan, have ever been publicly identified as Trust appointments. Anthropic has not said whether Trust-appointed directors still constitute a majority of the smaller board.

## Trustees

The founding trustees were named in the September 2023 announcement. Four of the five have since left, and Anthropic has documented each departure in a footnote to that announcement [1].

| Trustee | Role at appointment | Joined | Left |
|---|---|---|---|
| Neil Buddy Shah (Chair) | CEO, Clinton Health Access Initiative | 2023 | Serving |
| Jason Matheny | CEO, RAND Corporation | 2023 | December 2023, citing potential conflicts with RAND policy work |
| [Paul Christiano](/wiki/paul_christiano) | Founder, Alignment Research Center | 2023 | April 2024, to lead AI safety work at the US AI Safety Institute |
| Kanika Bahl | CEO and President, Evidence Action | 2023 | January 2026, to start the AI Access Initiative |
| Zach Robinson | Interim CEO, Effective Ventures US | 2023 | January 2026, to focus on nonprofit and philanthropic work |
| Richard Fontaine | CEO, Center for a New American Security | May 2025 | Serving |
| Mariano-Florentino (Tino) Cuellar | President, Carnegie Endowment for International Peace | January 2026 | Serving |
| Ben Bernanke | Distinguished Fellow, Brookings Institution | July 2026 | Serving |

Anthropic announced Fontaine's appointment in the first week of June 2025, a day after it released Claude Gov models built for US national security customers. TechCrunch reported the appointment on 2025-06-06 and Anthropic's own post carries a 2025-06-07 dateline. TechCrunch's Kyle Wiggers noted the timing and framed the Trust's safety role as something "Anthropic claims" rather than an independent finding [10]. Fontaine advised Senator John McCain on foreign policy, taught at Georgetown, and served four years on the Defense Policy Board [5][10]. Cuellar is a former California Supreme Court justice who led Stanford's Freeman Spogli Institute [11]. The role in the table above is the one he held when appointed: he stepped down as president of the Carnegie Endowment in July 2026 and returned to Stanford to lead the Center for Advanced Study in the Behavioral Sciences and to direct the Knight-Hennessy Scholars Program [11][24].

Ben Bernanke, chair of the Federal Reserve from 2006 to 2014 and a 2022 Nobel laureate in economics, joined on 2026-07-09 [12]. Anthropic quoted him saying that "the potential of artificial intelligence is enormous, and so is the range of outcomes. How that potential plays out will depend, in part, on the institutions we build around it" [12]. His appointment brings the Trust to four members against a design of five; Anthropic has not said when the fifth seat will be filled.

## Powers beyond electing directors

Anthropic describes the Trust as both selecting board members and "advising the Board and leadership on how the company can maximize the benefits of advanced AI and mitigate its risks" [5]. The Class T stock also carries what Anthropic calls "protective provisions" that require the Trust to receive notice of certain actions that could significantly alter the corporation or its business [1][4]. That is an information right, not a consent right: nothing published says the Trust can block such an action.

The advisory role has one concrete hook in a published document. The original [Responsible Scaling Policy](/wiki/responsible_scaling_policy), released alongside the Trust in September 2023, provided that changes to the policy "must be approved by the board following consultations with the Long Term Benefit Trust" [13]. That is a consultation requirement, not a veto.

Version 3.2 of the policy, effective 2026-04-29, extended the Trust's role: it "authorizes the LTBT to request external review of Risk Reports, gives the LTBT the power to approve our selection of external reviewers, and formalizes a requirement that we provide the LTBT with regular briefings" [14]. Approving the choice of external reviewers is the first published instance of the Trust holding a binding decision right over something other than board seats. Whether it has used the power is not public.

The policy has been revised twice since. The current version is 3.4, effective 2026-07-08, and it carries the Trust's role in its operative text rather than only in a change log: "In selecting external reviewers, we will consult with the Board and obtain the approval of the LTBT", and "Changes to the RSP will be proposed by the CEO and RSO, and approved by the Board in consultation with the LTBT" [14][23].

## The failsafe, and what has not been published

Anthropic disclosed from the start that the Trust can be overridden. The 2023 analysis describes three amendment routes: trustees plus stockholders acting together, trustees plus directors acting together before the Trust holds a board majority, and a supermajority stockholder vote without trustee consent, which the authors call "a kind of failsafe against the actions of the Voting Trustees" that "safeguards the interests of stockholders" [4]. Anthropic wrote that "the required supermajorities increase as the Trust's power phases in, on the theory that we'll have more experience, and less need for iteration, as time goes on, and the stakes will become higher" [1].

The same analysis is explicit about who can enforce the arrangement, and it is not the public. Under Delaware's purpose trust statute, "the Trust Agreement also authorizes the Trust to be enforced by the company and by groups of the company's stockholders who have held a sufficient percentage of the company's equity for a sufficient period of time" [4]. Neither the qualifying percentage nor the qualifying holding period has been published.

The numbers are the problem. Anthropic has never published the Trust Agreement or the supermajority percentages. The company also frames the design tension openly, acknowledging it is hard to make an arrangement resilient to end runs while it is also a first attempt that may need revision [1].

## Criticism

The most cited critique is Zach Stein-Perlman's May 2024 post "Maybe Anthropic's Long-Term Benefit Trust is powerless," written for AI Lab Watch [15]. His argument is narrow and about disclosure rather than motive: stockholders can apparently modify or abrogate the Trust by supermajority, the thresholds were never disclosed, and "it's impossible to assess this 'failsafe' without knowing the thresholds for these 'supermajorities.'" He argues that the available information is consistent with a Trust that is subordinate to stockholders and likely to lose its powers if it does something they dislike, and that Anthropic has not publicly demonstrated the Trust could do anything stockholders oppose [15].

Anthropic's general counsel Brian Israel told TIME that it would be impossible for Amazon and Google to force such a change, because the two companies hold no voting shares in Anthropic: they cannot elect directors, and their shares would not be counted in any supermajority required to rewrite the rules governing the Trust. Israel said the same of Anthropic's Series B stock, much of which was originally bought by the collapsed crypto exchange FTX [6]. The figure usually quoted alongside this point, that Amazon and Google each own less than 15 percent of Anthropic, came from a person familiar with the matter rather than from Israel, and it is not what his argument turns on: on his account the size of the stake is beside the point because the stock carries no vote [6].

Israel also described the PBC form as giving the board "a flexibility, not a mandate" [6], which is a concession rather than a defense: the public benefit purpose is broad enough to accommodate most decisions. He made a further concession about the Trust itself. Directors selected by the Trust owe the same fiduciary obligations to Anthropic's stockholders as any other director, which is why TIME concluded that "the LTBT ultimately has a limited influence on the company: while it will eventually have the power to select and remove a majority of board members, those members will in practice face similar incentives to the rest of the board" [6].

TIME's Billy Perrigo identified a structural gap that neither mechanism closes. "There is no obvious mechanism, however, for the public to sue Anthropic's board members for not pursuing its public benefit mission strongly enough" [6]. Harvard Law professor Noah Feldman, who consulted on the design, told TIME that "nothing exactly like this has been tried, to my knowledge" [6]. Daniel Colson of the AI Policy Institute argued in the same piece that private governance structures are not a substitute for government regulation of frontier [AI safety](/wiki/ai_safety) risk [6].

A separate line of criticism concerns follow-through. Journalist Garrison Lovely reported in May 2025 that Anthropic had dropped its original commitment to define [ASL-4](/wiki/ai_safety_levels) before deploying an ASL-3 model, and used the Trust's consultation role in the Responsible Scaling Policy to argue that the body had not visibly prevented the change [16]. Lovely also noted the mismatch between the reported November 2024 majority milestone and the small number of directors Anthropic then listed publicly [16].

## Has it been tested?

Not publicly, in the sense that matters. The Trust has appointed directors, which is the exercise of its ordinary power, and Anthropic says it reached a board majority in April 2026 [3]. But no reported episode involves the Trust removing a director, blocking a decision, or overruling management, and no reported episode involves stockholders invoking the supermajority failsafe.

The closest real-world stress test is indirect. In February 2026 Anthropic refused a Defense Department demand for unrestricted military use of Claude, holding out for limits on mass surveillance of Americans and on fully autonomous targeting. The Pentagon's ultimatum was to grant unrestricted use or face a ban from all government contracts; no agreement was reached, and CBS News reported that President Trump and Defense Secretary Pete Hegseth followed through [17]. Writing in Bloomberg Law on 2026-04-22, Ofer Eldar of UC Berkeley and Mark Orberg of Copenhagen Business School argued that Anthropic's design made that refusal easier to sustain, because a company whose board is majority-appointed by a purpose trust cannot as readily reframe a contested contract as a straightforward revenue opportunity. They added that the supermajority failsafe "appears reserved for extreme circumstances, not for directing ordinary business strategy" [18]. That is an argument about the structure shaping incentives, not evidence that the Trust used a formal power. Anthropic has not said the Trust was involved in the Pentagon decision, and contemporaneous reporting on the dispute did not mention the Trust or the board: CBS framed the fight as "a question of who controls how artificial intelligence models are used, the Pentagon or the company's CEO" [17].

## Comparison with OpenAI

[OpenAI](/wiki/openai) is the obvious reference point, and the two structures differ in the identity of the controller rather than the general idea. OpenAI completed its own [restructuring](/wiki/openai_pbc_restructuring) on 2025-10-28: the original nonprofit became the OpenAI Foundation and the for-profit arm became OpenAI Group PBC, with the Foundation controlling the PBC and holding a 26 percent equity stake, worth about $130 billion at the time, against Microsoft's roughly 27 percent on an as-converted diluted basis [19][22]. OpenAI says the recapitalization was completed "after nearly a year of engaging in constructive dialogue with the offices of the Attorneys General of California and Delaware" [19].

Three differences matter for evaluating the Trust. First, OpenAI's controller is a charitable nonprofit with a large economic stake, while Anthropic's is a purpose trust with a near-zero economic stake. Second, OpenAI's controller sits under the supervision of two state attorneys general who can act on charitable-trust grounds; Anthropic's Trust is enforceable only by the company and by qualifying stockholder groups, with no comparable public enforcer, which is the gap Perrigo described [4][6]. Third, OpenAI's nonprofit board demonstrated in November 2023 what unchecked controller power looks like when it removed and then reinstated Sam Altman. Anthropic's design deliberately avoids concentrating that kind of abrupt authority, at the cost of the Trust having correspondingly weaker levers [6].

## The IPO question

Anthropic confidentially submitted a draft Form S-1 to the SEC on 2026-06-01, after closing a $65 billion Series H at a $965 billion post-money valuation on 2026-05-28 [20][21]. Neither announcement mentioned the Trust or said how the governance structure would carry into a public listing; the S-1 notice identifies the issuer as "Anthropic, PBC" and says nothing further about governance [20][21]. Because the draft registration statement is confidential, its risk factors and governance disclosures are not public. A [potential listing](/wiki/anthropic_ipo_2026) therefore raises questions that currently have no verifiable answers.

The open questions are concrete. Public-market investors normally expect to be able to replace directors through proxy voting; a structure in which an unelected body appoints a majority of the board removes that lever for the seats it controls. Exchange listing standards and index inclusion rules treat controlled companies and multi-class structures differently from ordinary ones. It is also unclear whether the supermajority failsafe would be recalculated against a public float, and whether the escalating Class T rights would continue on the same schedule after a listing. As of 2026-07-27, Anthropic has published nothing on any of these points, and no primary-source reporting resolves them. Commentary circulating on the subject is largely speculation built on the same public facts, and this article does not repeat it.

## See also

- [Anthropic](/wiki/anthropic)
- [Responsible Scaling Policy](/wiki/responsible_scaling_policy)
- [AI governance](/wiki/ai_governance)
- [OpenAI restructuring](/wiki/openai_pbc_restructuring)
- [Anthropic IPO](/wiki/anthropic_ipo_2026)
- [Dario Amodei](/wiki/dario_amodei), [Daniela Amodei](/wiki/daniela_amodei)
- [Centre for the Governance of AI](/wiki/govai)
- [Compute governance](/wiki/compute_governance)
- [The Anthropic Institute](/wiki/anthropic_institute)

## References

1. "The Long-Term Benefit Trust." Anthropic, 2023-09-19. https://www.anthropic.com/news/the-long-term-benefit-trust
2. "Company." Anthropic, accessed 2026-07-27. https://www.anthropic.com/company
3. "Anthropic's Long-Term Benefit Trust appoints Vas Narasimhan to Board of Directors." Anthropic, 2026-04-14. https://www.anthropic.com/news/narasimhan-board
4. Morley, John, David Berger and Amy Simmerman. "Anthropic Long-Term Benefit Trust." Harvard Law School Forum on Corporate Governance, 2023-10-28. https://corpgov.law.harvard.edu/2023/10/28/anthropic-long-term-benefit-trust/
5. "National security expert Richard Fontaine appointed to Anthropic's Long-Term Benefit Trust." Anthropic, dated 2025-06-07. https://www.anthropic.com/news/national-security-expert-richard-fontaine-appointed-to-anthropic-s-long-term-benefit-trust
6. Perrigo, Billy. "How Anthropic Designed Itself to Avoid OpenAI's Mistakes." TIME, 2024-05-30 (updated 2026-01-19). https://time.com/6983420/anthropic-structure-openai-incentives/
7. Buntz, Brian. "Anthropic's oversight trust just hit majority control." R&D World, 2026-04-15. https://www.rdworldonline.com/anthropics-oversight-trust-just-hit-majority-control-the-tipping-point-was-adding-novartis-ceo-vas-narasimhan-to-its-board/
8. "Reed Hastings appointed to Anthropic's board of directors." Anthropic, 2025-05-28. https://www.anthropic.com/news/reed-hastings
9. "Anthropic: Former Trump aide Liddell on board of directors." CNBC, 2026-02-13. https://www.cnbc.com/2026/02/13/anthropic-ai-chris-liddell-microsoft-trump-board.html
10. Wiggers, Kyle. "Anthropic appoints a national security expert to its governing trust." TechCrunch, 2025-06-06. https://techcrunch.com/2025/06/06/anthropic-appoints-a-national-security-expert-to-its-governing-trust/
11. "Mariano-Florentino Cuellar appointed to Anthropic's Long-Term Benefit Trust." Anthropic, 2026-01-21. https://www.anthropic.com/news/mariano-florentino-long-term-benefit-trust
12. "Ben Bernanke appointed to Anthropic's Long-Term Benefit Trust." Anthropic, 2026-07-09. https://www.anthropic.com/news/ben-bernanke
13. "Anthropic's Responsible Scaling Policy." Anthropic, 2023-09-19. https://www.anthropic.com/news/anthropics-responsible-scaling-policy
14. "Anthropic's Responsible Scaling Policy" (version history and change log; current version 3.4, page last updated 2026-07-08). Anthropic. https://www.anthropic.com/responsible-scaling-policy
15. Stein-Perlman, Zach. "Maybe Anthropic's Long-Term Benefit Trust is powerless." AI Lab Watch, 2024-05-27. https://ailabwatch.org/blog/anthropic-trust-powerless
16. Lovely, Garrison. "Exclusive: Anthropic is Quietly Backpedalling on its Safety Commitments." Obsolete, 2025-05-22. https://www.obsolete.pub/p/exclusive-anthropic-is-quietly-backpedalling
17. Kent, Jo Ling and Caitlin Yilek. "What's behind the Anthropic-Pentagon feud." CBS News, 2026-02-25. https://www.cbsnews.com/news/anthropic-pentagon-pete-hegseth-feud/
18. Eldar, Ofer and Mark Orberg. "Anthropic, Pentagon Standoff Shows Why AI Company Design Matters." Bloomberg Law, 2026-04-22. https://news.bloomberglaw.com/legal-exchange-insights-and-commentary/anthropic-pentagon-standoff-shows-why-ai-company-design-matters
19. "Our structure." OpenAI, accessed 2026-07-27. https://openai.com/our-structure/
20. "Anthropic confidentially submits draft S-1 to the SEC." Anthropic, 2026-06-01. https://www.anthropic.com/news/confidential-draft-s1-sec
21. "Anthropic raises $65B in Series H funding at $965B post-money valuation." Anthropic, 2026-05-28. https://www.anthropic.com/news/series-h
22. "The next chapter of the Microsoft-OpenAI partnership." Microsoft, 2025-10-28. https://blogs.microsoft.com/blog/2025/10/28/the-next-chapter-of-the-microsoft-openai-partnership/
23. "Responsible Scaling Policy," version 3.4 (PDF), effective 2026-07-08. Anthropic. https://cdn.sanity.io/files/4zrzovbb/website/0bacdc8440ea96e62a8766d99ebe1d4eea6d5f3a.pdf
24. "Tino Cuellar to Step Down as Carnegie Endowment President in July 2026." Carnegie Endowment for International Peace, 2026-01-15. https://carnegieendowment.org/posts/2026/01/tino-cuellar-to-step-down-as-carnegie-endowment-president-in-july-2026
25. "Jay Kreps appointed to Anthropic's Board of Directors." Anthropic, 2024-05-29. https://www.anthropic.com/news/jay-kreps-appointed-to-board-of-directors