Corporate Governance and the Long-Term Benefit TrustWho is a corporation for? This question has been asked repeatedly since the joint-stock company was first invented. The standard answer is: shareholders. Shareholders elect directors, directors select the CEO, and the CEO manages employees. In this chain, the ultimate beneficiary is the shareholder, and the corporation exists to maximize profit. The rise of artificial intelligence is shaking that premise to its foundations.Anthropic is a company attempting to give that question a structural answer. Its Long-Term Benefit Trust (LTBT) is an effort to redesign corporate governance itself, so that AI remains genuinely beneficial to humanity. Yet the attempt has attracted both admiration and skepticism in equal measure. The design intent is noble; the structural reality is far more complicated. This article examines how the LTBT works, where it falls short, how it compares to Japan's third-party committee system, what global precedents exist, and what lies at the heart of the problem that few dare to discuss openly.What Is the LTBT?The LTBT is an independent governance body established by Anthropic in 2023. It comprises five trustees, none of whom hold any financial stake in Anthropic. They have the authority to elect an increasing share of Anthropic's board of directors, with a majority to be achieved within four years. Members are drawn from the fields of AI safety, national security, public policy, and social enterprise.Its legal form is a Delaware "purpose trust" — a rare structure that exists not to enrich specific beneficiaries but to advance a defined purpose. That purpose is "the responsible development and maintenance of advanced AI for the long-term benefit of humanity," language that is also enshrined in Anthropic's corporate charter.Following its Series C funding round, Anthropic amended its charter to create a new class of shares — Class T Common Stock — held exclusively by the LTBT. This special class is the mechanism by which trustees gain the power to elect directors. The arrangement also includes protective provisions requiring Anthropic to notify the LTBT before taking certain significant corporate actions. The LTBT additionally receives quarterly reports on AI safety and timely updates on model evaluation results tied to release decisions. This creates a structural reduction in information asymmetry, functioning as an ongoing monitoring mechanism even in the absence of direct intervention.The most fundamental difference from conventional corporate governance is the direction of power. Normally, shareholders sit at the top, controlling the CEO through the board. At Anthropic, an external independent body selects the majority of the board, making mission-first behavior structurally embedded rather than dependent on goodwill. Even the founders — Dario Amodei and Daniela Amodei — are bound by this arrangement.How It Differs From Japan's Third-Party CommitteeMany readers familiar with Japanese corporate culture will immediately think of the daisansha iinkai — the third-party committee — and ask whether the LTBT is simply a version of the same thing. It is not. Understanding precisely how they differ is the quickest path to grasping the LTBT's true nature.Japan's third-party committee emerged in 1997 following the collapse of Yamaichi Securities, and became widely established after the Japan Federation of Bar Associations issued its guidelines on the subject in 2010. It has since been deployed in response to misconduct at organizations ranging from the Takarazuka Revue (harassment), Fuji Television, Toshiba (accounting fraud), and Nihon University — appearing, in short, whenever a scandal erupts. Lawyers and academics investigate the facts, analyze causes, and publish reports with recommendations for preventing recurrence. It is, in essence, a reactive crisis-management tool.The contrast with the LTBT is stark across three dimensions: purpose, timing, and authority.On purpose: the third-party committee exists to establish the truth after something has gone wrong. It is a retrospective prosecutor. The LTBT, by contrast, is a standing governance institution that monitors management and appoints directors before anything goes wrong. The difference is that between a fire station and a fire engine. The committee is called when the building is already burning; the LTBT is meant to prevent the fire from starting.On timing: the third-party committee is established by the company itself, after a problem has surfaced. The company decides whether to create one — which means a company can simply choose not to, if inconvenient. The LTBT is a permanent body embedded at the company's founding. Dissolving it requires a shareholder supermajority, a high and deliberate bar.On authority: the third-party committee holds no legal compulsory investigative powers. Interviews and document disclosures depend entirely on voluntary cooperation. Management can stonewall, withhold documents, or restrict the scope of investigation — all of which happened at Toshiba, where the committee was reported to have excluded the very issues that mattered most, at the apparent direction of executives. Once the report is published, there is nothing legally forcing the company to follow any of its recommendations.The LTBT, by contrast, holds the power to elect and remove directors — the structural core of any corporation. It does not merely advise. It has the institutional capacity to replace a CEO or director who fails to act in accordance with the mission.The reason Japan's third-party committees are so often dismissed as rubber-stamp exercises — otesori, or "self-serving investigations" — comes down to exactly this. The company appoints the members and pays their fees. The suspicion that conclusions will tilt toward the company is structurally baked in. Research by one accounting scholar found that, at the rate of roughly one new committee per five days in Japan, only a small fraction of reports meet an acceptable standard — and many are judged "pointless or actively harmful."To be fair, the LTBT faces analogous criticisms: the trust agreement is not public, activity has been limited, and expert trustees have departed. The problem of a governance structure that looks credible from the outside but lacks substance within is not unique to Japan. It appears to be a universal one.To summarize: the third-party committee is reactive, temporary, and toothless. The LTBT is permanent, pre-emptive, and — in theory — powerful. For a Japanese company to have something equivalent to the LTBT, it would need to look not to the third-party committee model, but to the strengthening of independent outside directors or conversion to a three-committee governance structure. The two institutions operate at entirely different levels.The Uncomfortable Reality: Criticism and LimitsHaving established the LTBT's structural advantages, its critics deserve a full hearing.The greatest problem is the inversion of enforcement authority. The trust agreement can be amended or revoked by a shareholder supermajority without trustee consent. More troublingly, the legal power to enforce the trust agreement rests not with the trustees themselves, but with shareholders holding a sufficient percentage of equity over a sufficient period. In practice, this means that if a trustee objects to a course of action as contrary to humanity's long-term interests, shareholders — not trustees — hold the legal tools to override that judgment. The watchdog is structurally constrained by the very parties it is meant to watch. This inverted architecture is the LTBT's most fundamental contradiction.The trust agreement itself has never been made public. The supermajority threshold — how large a shareholder coalition is needed to override the LTBT — is undisclosed. Given that Amazon and Google are among Anthropic's largest investors, there is a genuine question of how easily they might together reach that threshold. The critics' challenge — "if the terms are truly robust, why not publish them?" — has received no clear answer. A lack of transparency and a lack of trustworthiness tend to travel together.The LTBT's track record is also thin. It was designed to appoint up to five board members — a majority — yet has appointed only one. Its own membership has fallen from five to three. None of the current trustees possess deep technical expertise in AI. Paul Christiano, one of the world's leading AI safety researchers, left to lead the US AI Safety Institute. Jason Matheny, former CEO of the RAND Corporation, departed to avoid potential conflicts of interest. The two people who understood AI risk most deeply chose to leave. That is not easy to dismiss.Whether the LTBT can even remove a CEO remains unclear. Reporting has suggested that, depending on the contents of the undisclosed investor rights agreement, even LTBT-appointed directors may lack the practical ability to fire the CEO. The board's greatest lever of power is its authority to replace leadership. If that lever is blocked, the oversight function is largely ceremonial. Anthropic's own admission that the LTBT is "an experiment" — combined with the spectacle of OpenAI's nonprofit board failing to hold its CEO accountable in 2023 — does nothing to quiet these concerns.Global Precedents: A Hundred Years of EvidenceDenmark offers the most instructive comparison. Roughly a quarter of Denmark's top 100 companies are foundation-owned. The Carlsberg Foundation was established in 1876 and has maintained majority voting control for over 130 years. The Novo Nordisk Foundation held net assets of approximately $220 billion as of February 2026, making it the world's largest charitable foundation. Maersk's articles of association explicitly prohibit the foundation from ever selling its shares. What these companies share is an absolute lock: foundations hold majority voting rights, and those shares cannot be divested.The practical consequence is profound. Shielded from short-term shareholder pressure, these companies operate on timescales measured in decades rather than quarters. It is also why Danish companies have largely avoided the hostile takeovers and cross-border mergers that have reshaped competitors elsewhere.Patagonia transplanted this Nordic model to the United States in 2022. Founder Yvon Chouinard transferred 100% of voting shares to the Patagonia Purpose Trust and the remaining 98% to the Holdfast Collective, an environmental nonprofit. All profits — roughly $1 billion per year — are directed to environmental causes. Cumulative donations since the 2022 transition now exceed $1 billion. In Europe, Germany's Robert Bosch Foundation, established in 1964, holds approximately 92% of voting rights, and IKEA has operated under a Dutch foundation structure since 1982.Compared to these models, the LTBT's design — gradual, amendable, reversible — falls well short as a mission lock. But context matters. The Danish foundation model was not born complete either. Carlsberg did not fully transition to foundation control until 1888, twelve years after the foundation was created. Maersk did not establish its foundation until 1953, nearly fifty years after the company was founded. Seen against that historical sweep, a four-year phase-in timeline is not inherently defeatist. The question is whether it will be gutted before the phase-in is complete.What No One Is SayingIt is worth asking whether the LTBT functions, in practice, primarily as a signaling device — a way of reassuring both regulators and the public that Anthropic will not behave like an ordinary corporation. Anthropic's brand is built on safety. The LTBT reinforces that brand. It reduces regulatory risk, improves talent acquisition, and reassures investors that the company will not be destroyed by its own recklessness. Seen this way, the LTBT is not in tension with Anthropic's commercial strategy — it is an integral part of it. Whether this represents "strategic alignment with mission" or "sophisticated reputational management" depends on your reading of the company's intentions.The successive departures of the two AI safety experts are harder to explain away. Both Christiano and Matheny knew the full contents of the trust agreement — documents the public has never seen. Christiano's move to a government body could reflect a judgment that public regulatory frameworks offer more genuine leverage over AI development than a private trust. Or it could reflect something he encountered within the LTBT itself. Either way, the fact that the person who understood AI risk most rigorously chose to work outside the structure is not a trivial data point.The deeper issue is one the Danish model clarifies by contrast. Foundation ownership has lasted a century because the foundation cannot sell the shares. Full stop. No exceptions. The legal constraint is absolute. The LTBT contains no such absolute. The agreement can be amended; the shares can be transferred; the threshold for doing so is secret. Ultimately, whether the LTBT functions as intended depends less on its legal architecture than on whether Anthropic's founders and major shareholders genuinely prioritize mission over money when those two things conflict. No governance document can substitute for human integrity.When Anthropic says the LTBT is "an experiment," those words carry real weight. The verdict will come — if it comes at all — when humanity and AI reach their first genuine crisis point. By the time we know whether the experiment succeeded or failed, the window for correction may already have closed. That is the deepest dilemma of this governance experiment, and it is one no structural innovation has yet learned to solve.ReferencesAnthropic. "The Long-Term Benefit Trust." Anthropic News, September 19, 2023. https://www.anthropic.com/news/the-long-term-benefit-trustAnthropic. "National Security Expert Richard Fontaine Appointed to Anthropic's Long-Term Benefit Trust." June 2025. https://www.anthropic.com/news/national-security-expert-richard-fontaine-appointed-to-anthropic-s-long-term-benefit-trustMorley, John, David J. Berger, and Amy L. Simmerman. "Anthropic Long-Term Benefit Trust." Harvard Law School Forum on Corporate Governance, October 28, 2023. https://corpgov.law.harvard.edu/2023/10/28/anthropic-long-term-benefit-trust/"Maybe Anthropic's Long-Term Benefit Trust is Powerless." LessWrong / AI Lab Watch, May 2024. https://www.lesswrong.com/posts/sdCcsTt9hRpbX6obP/maybe-anthropic-s-long-term-benefit-trust-is-powerless"Anthropic Long-Term Benefit Trust." Longterm Wiki, March 2026. https://www.longtermwiki.com/wiki/E407Hatta, Shinji. "The Deception of Third-Party Committees." Toyo Keizai Shinposha. [In Japanese]Japan Federation of Bar Associations. "Guidelines on Third-Party Committees for Corporate Misconduct." 2010. [In Japanese]"On the Utility and Limits of Third-Party Committees." Japan Association of Corporate Directors, 2025. https://www.jacd.jp/news/column/column-opinion/251110_post-359.html"The Effectiveness and Limits of Third-Party Committees in Corporate Misconduct Response." Note, 2026. https://note.com/sane_viola893/n/nedc0d299547f"Rethinking Ownership: Putting Purpose at the Center." Harvard ALI Social Impact Review, 2024. https://www.sir.advancedleadership.harvard.edu/articles/rethinking-ownership-putting-purpose-at-center"Denmark's Foundation Model: What Is It?" Bloomberg, November 13, 2025.Novo Nordisk Foundation. "Ownership." https://novonordiskfonden.dk/en/who-we-are/ownership/"Governance of Foundation-Owned Firms." Journal of Legal Analysis, Oxford Academic. https://academic.oup.com/jla/article/13/1/172/6180587"Elon Musk's xAI Secretly Dropped Its Benefit Corporation Status While Fighting OpenAI." CNBC, August 25, 2025.OpenAI. "Evolving OpenAI's Structure." https://openai.com/index/evolving-our-structure/"Unless Its Governance Changes, Anthropic Is Untrustworthy." EA Forum, December 2025. https://forum.effectivealtruism.org/posts/6XbtL93kSFJwX45X2