Anthropic 的精巧架构：如何将华尔街资本力量拒之门外

Anthropic 想要鱼与熊掌兼得。随着即将到来的 IPO，它希望利用公开市场的巨额资本，同时坚称绝不会在安全且造福人类地开发 AI 这一核心使命上妥协。但更强大的模型能力意味着更大的风险，而渴望回报的股东们，可能没有多少耐心容忍那些延缓产品上市进程的昂贵安全措施。

Anthropic 的头号竞争对手 OpenAI 的 CEO Sam Altman 曾援引投资者的压力作为该公司今年不寻求 IPO 的原因之一。而 Anthropic 则希望借助一套令人惊叹的治理架构来抵御投资者的压力。

一种“此前从未被使用过的独特而复杂的治理架构……专为这家创始人主导、使命驱动的公司而创新设计”。

宾夕法尼亚大学法学教授 Elizabeth Pollman 告诉 The Verge，Anthropic 已提议采用一种“此前从未被使用过的独特而复杂的治理架构”上市，该架构是“专为这家创始人主导、使命驱动的公司而创新设计的”。专家表示，这应该能给 Anthropic 追求利润的自由，同时避免股东强迫其在安全上偷工减料，或在系统未经充分测试之前就仓促推向市场。但即便最强有力的法律保护也是有限的，没有任何企业架构——无论设计得多巧妙——能够彻底将一家上市公司与市场需求隔绝开来。

首先，Anthropic 是一家公共利益公司（Public Benefit Corporation，简称 PBC）。简而言之，它在法律上被要求不能仅仅专注于股东回报最大化，还必须将商业利益与自身声明的社会使命以及员工、特定社群等利益相关者的权益相平衡。Patagonia 和 Kickstarter 是其中较为知名的代表，两者的使命分别聚焦于环境保护和支持创意群体。

另一项关键防线旨在通过限制股东对公司最高权力杠杆之一——董事会的影响力，来保护 Anthropic 的使命。董事负责监督公司并任命包括 CEO 在内的管理层。董事也可以同时担任高管职务，Anthropic 正是如此：CEO Dario Amodei 和总裁 Daniela Amodei 均在董事会任职。

股东通常对董事会成员拥有相当大的影响力，这为他们提供了在对公司运营不满时更换董事或推动变革的手段。例如，如果股东认为由于非强制性的安全测试导致产品推迟上市是不必要的拖延并对此感到不满，他们就可以罢免董事，换上更符合自身利益（即更少谨慎、更多赚钱）的人选。

“从普通股东无权选拔董事或做出商业决策的角度来看，Anthropic 完全隔绝了追求利润的压力。”

在 Anthropic 拟议的治理架构下，普通股东在董事会人选上几乎没有任何发言权。德克萨斯大学法学院商法教授 Jens Dammann 解释称，董事会 7 名董事中的 4 名将由一个名为“长期利益信托”（Long-Term Benefit Trust）的特别机构选出。这对于一家大型科技公司来说是非常罕见的设置。

该信托由“五名在经济上无利益关系的投票受托人”组成，他们在“AI 安全、国家安全、公共政策和公益企业”领域拥有专业背景。通过一种名为 T 类股票（Class T stock）的特殊股票类别，受托人有权任命或罢免董事，以践行其既定宗旨：确保“Anthropic 负责任地平衡股东的财务利益与受 Anthropic 行为及其公共利益宗旨影响者的利益”。该信托旨在独立于公司的商业压力运作，在理论上确保 Anthropic 的使命始终处于第一位。

普通股东从技术上讲对董事会剩余的 3 个席位拥有发言权。然而，Anthropic 还设立了另一道防线，将权力牢牢掌握在创始人手中——即名为“创始人有限责任公司”（Founder LLC）的实体，它将在股东事务中控制 50.1% 的投票权。鉴于许多股东投票都是由简单多数决定的，Founder LLC 基本上赋予了 Anthropic 创始人阻止或强推决策的权力，无论其他股东如何投票。哥伦比亚大学法学院教授 Dorothy Lund 告诉 The Verge，这意味着他们实际上将敲定剩余的 3 名董事。

招股书显示，Founder LLC “最初”将由 Anthropic 的 7 位创始人组成：Dario Amodei、Daniela Amodei、Tom Brown、Jack Clark、Jared Kaplan、Sam McCandlish 以及 Christopher Olah。“我们的 7 位联合创始人最初将是唯一成员。”正如 The San Francisco Standard 所报道的那样，另一位联合创始人 Ben Mann（他共同领导公司的实验性业务 Anthropic Labs）引人注目地缺席了该名单。

Dammann 告诉 The Verge，总而言之，这些安排将赋予 Anthropic 创始人及信托对公司方向的“巨大权力”，保护其免受作为上市公司的许多外部压力。“鉴于 Anthropic 采用的非同寻常的治理架构，我认为上市本身并不会削弱其将安全置于商业利益之上的能力。”

科罗拉多大学法学院教授 Ann Lipton 对此表示赞同，她向 The Verge 表示，从法律层面上讲，“Anthropic 完全隔绝了追求利润的压力，因为普通股东完全没有选拔董事或做出商业决策的能力。”但 Lipton 提醒道，创始人本身在公司持有经济利益，这可能会影响他们的决断。

“只要公司运转良好，投资者可能不在乎自己没有话语权。但一旦情况恶化，他们可能会变得相当沮丧。”

此外，法律保护与免疫市场压力之间存在很大差距。Lipton 指出，Anthropic 与所有 AI 公司一样，“极度渴求资本”。“我并不清楚是否有任何治理架构能够抵御这种市场现实：除非 Anthropic 能以比竞争对手更快的节奏捕捉商业变现机会，否则它将无法维持自身运营的资金消耗。”Lund 也表达了类似的担忧。她表示，Anthropic 仍需“保持与投资者的良好关系”，并补充道“竞争压力和资本需求可能会随着时间推移侵蚀对安全的承诺”。针对 The Verge 就其是否真能抵挡股东要求以牺牲安全为代价大力加速 AI 发展的压力所提出的置评请求，Anthropic 未予回应。

如果 Anthropic 的创始人开始离职，另一个潜在漏洞就会显现。公司在招股书中表示，其成功“高度依赖”核心团队（尤其是创始人）的持续任职。Dario 和 Daniela 被特别强调为推进公司研究与使命的“至关重要”的人物。但除了管理层流失之外，他们的离开还可能引发公司架构的重大变动，进而可能拆除部分防止 Anthropic 追逐利润而非安全的防护栏。

在拟议的架构下，如果 Founder LLC 的多数成员离职，或者该群体选择解散，Founder LLC 将宣告“终止”（sunset）。Dammann 告诉 The Verge，一旦发生这种情况，普通股东可能会获得削弱信托保护措施的权力，即便这违背受托人的意愿，且得不到持有不同类别股票的创始人或员工的支持。他解释称，颠覆该信托的其他途径同样存在，包括公众股东、员工持股者与创始人之间的协同行动。

目前来看，Anthropic 似乎有可能兼顾二者。但这些旨在守护其 AI 安全承诺的防线远非坚不可摧，很大程度上取决于市场压力、创始人是否留任以及信托的导向。Dammann 表示，他看到了 Anthropic 的使命与投资者诉求之间存在着“无法完全圆满解决的根本性困境”。只要一切顺风顺水，这种张力或许尚可维持。“只要一切顺利，投资者可能不会在意自己失去话语权，”Dammann 说道，“但一旦局势逆转，他们可能会变得相当沮丧。”

Anthropic wants to have it both ways. With its impending IPO, it wants to tap the vast riches of the public markets and insists it doesn’t want to compromise its core mission of developing AI safely and for the benefit of humanity. But greater model capabilities bring greater risks, and shareholders eager for returns may have little patience for costly safety measures that slow products on their way to market.

Sam Altman, the CEO of Anthropic’s archrival, OpenAI, has cited investor pressures as one reason the company is not pursuing an IPO this year. Anthropic hopes it can resist investor pressures with a mind-boggling governance structure.

A “unique and complicated governance structure that has not been used before…creatively designed for this founder-led and mission-driven company.”

University of Pennsylvania law professor Elizabeth Pollman told The Verge that Anthropic has proposed going public with a “unique and complicated governance structure that has not been used before” and that has been “creatively designed for this founder-led and mission-driven company.” Experts say it should give Anthropic freedom to pursue profits without letting shareholders force it to cut corners on safety or rush systems out before they’re adequately tested. But even the strongest legal protections can only go so far, and no corporate structure, no matter how well designed, can completely insulate a public company from the demands of the market.

Foremost, Anthropic is a public benefit corporation. Simply put, it is legally required to focus on more than just maximizing shareholder returns, balancing these with its stated social mission and the interests of stakeholders, such as employees and specific communities. Patagonia and Kickstarter are among the better-known examples, with missions centered on environmental conservation and supporting creative communities, respectively.

Another key safeguard aims to protect Anthropic’s mission by limiting shareholder influence over one of the biggest levers of corporate power: the company’s board of directors. Directors oversee the company and appoint its leadership, including the CEO. Directors can also hold executive roles, as is the case with Anthropic, with CEO Dario Amodei and president Daniela Amodei both sitting on the board.

Shareholders typically have considerable sway over who sits on the board, giving them a means to replace directors or push for changes if they are unhappy with how the company is being run. For example, if shareholders are unhappy at what they perceive as unnecessary delays to market for optional safety testing, they may remove directors and replace them with ones more aligned with their interests, such as making money and being less cautious.

“Anthropic is entirely insulated from profit seeking pressures, in the sense that ordinary shareholders have no ability to select the board or make business decisions.”

Under Anthropic’s proposed governance structure, ordinary shareholders won’t have much of a say over who is on the board at all. Instead, four of the board’s seven directors will be chosen by a special body called the long-term benefit trust, explains Jens Dammann, a professor in business and commercial law at the University of Texas School of Law. It is a highly unusual setup for a major technology company.

The trust is made up of “five financially disinterested voting trustees” with expertise in “AI safety, national security, public policy, and social enterprise.” Through a special class of share called Class T stock, the trustees can appoint or remove directors in pursuit of their stated purpose: ensuring “that Anthropic responsibly balances the financial interests of stockholders with the interests of those affected by Anthropic’s conduct and our public benefit purpose.” The trust is designed to operate independently of the company’s commercial pressures, allowing it, in theory, to put Anthropic’s mission first.

Ordinary shareholders technically have a say in the board’s three remaining seats. However, Anthropic has created another safeguard that keeps power firmly in the hands of its founders, called “Founder LLC,” which will control 50.1 percent of the voting power on shareholder matters. Given that many shareholder votes are decided by a simple majority, Founder LLC basically gives Anthropic’s founders the ability to block or force through decisions, regardless of how other shareholders vote. This means that they will effectively choose the remaining three directors, Columbia Law School professor Dorothy Lund told The Verg e.

The prospectus says Founder LLC will “initially” be seven of Anthropic’s founders: Dario Amodei, Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, and Christopher Olah. “Our seven co-founders will initially be the sole members.” Notably absent is another cofounder, Ben Mann, who co-leads the company’s experimental Anthropic Labs arm, as reported by The San Francisco Standard.

All in all, the arrangements would give Anthropic’s founders and the trust “tremendous power” over the company’s direction, shielding it from many of “the pressures associated with being a public company,” Dammann told The Verge. “Given the unusual governance structure Anthropic has chosen, I don’t see how going public, by itself, would undermine its ability to prioritize safety over commercial interests.”

Ann Lipton, a professor at the University of Colorado Law School, agreed, telling The Verge that, legally speaking, “Anthropic is entirely insulated from profit seeking pressures, in the sense that ordinary shareholders have no ability to select the board or make business decisions.” But Lipton cautioned that the founders themselves have financial stakes in the company, which could influence their decisions.

“Investors may not mind their lack of voice as long as things go well. But if things go sour, they may become quite frustrated.”

And there’s a big difference between legal protections and immunity from market pressures. Anthropic, like all AI companies, “desperately needs capital,” Lipton said. “It’s not clear to me that any governance structure will protect against the market reality that unless Anthropic takes advantage of profit making opportunities at a faster pace than its competitors, it will not be able to fund its operations.” Lund expressed similar concerns. Anthropic will still “need to stay in the good graces of their investors,” she said, adding that “competitive pressure and capital needs can erode safety commitments over time.” Anthropic did not respond to The Verge ’s request for comment on whether it felt the company could actually resist shareholder pressure to massively accelerate AI at the cost of safety.

Another potential vulnerability emerges if Anthropic’s founders start leaving. In its prospectus, the company says its success is “highly dependent on the continued service” of its senior team, particularly its founders. Dario and Daniela are singled out as being especially “critical” to advancing the company’s research and mission. But beyond the loss of leadership, their departure could trigger significant changes in how the company is structured, potentially stripping away some of the safeguards preventing Anthropic from pursuing profit over safety.

Under the proposed structure, Founder LLC would “sunset” if a majority of its members leave or the group chooses to dissolve it. Once that happens, ordinary shareholders could gain the power to weaken the trust’s protections, even against the wishes of the trustees, and without the support of the founders or employees, who hold different classes of stock, Dammann told The Verge. Other means of subverting the trust also exist, he explained, including coordinated action by the public shareholders, employee stockholders, and founders.

For now, it seems, Anthropic can probably have it both ways. But the protections designed to preserve its commitment to AI safety are far from ironclad, and much depends on market pressures, whether its founders stick around, and directions of the trust. Dammann said he sees a “fundamental dilemma that cannot be resolved in a fully satisfactory way” between Anthropic’s mission and investor demands. As long as things are going well, this tension may be manageable. “Investors may not mind their lack of voice as long as things go well,” Dammann said. “But if things go sour, they may become quite frustrated.”

Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates.

• Robert Hart

• AI

• Anthropic

• Policy

• Report