Anthropic 在自身 IPO 文件中警告 AI 存在“灾难性”风险

随着 Anthropic 筹备备受期待的上市，据报道，该公司的 IPO 文件预览披露了其不断扩大的亏损、管理层试图保留控制权的方案，以及其 AI 开发计划如何可能“进一步增加我们的模型造成伤害的风险”。这些披露出现之际，Anthropic 正瞄准 2 万亿美元估值，超过其四个月前 9650 亿美元估值的两倍，这使其有望超过 SpaceX，成为史上规模最大的 IPO。

据 Reuters 报道，该媒体审阅了 Anthropic 的招股说明书。这家 AI 初创公司计划在未来几年投入 5180 亿美元，用于云服务、计算及基础设施相关义务，押注人工智能将成为全球经济不可或缺的一部分。尽管 Anthropic 的收入在 2025 年增长 12 倍、达到近 46 亿美元，但该公司同年报告净亏损 420 亿美元，仅经营活动就亏损逾 80 亿美元。

Anthropic 主要通过其 Claude 模型的按量计费 Token 使用量及客户订阅赚钱——这与 OpenAI、Google 等竞争供应商的商业模式类似——但 Anthropic 的收入来源似乎较为有限。Financial Times 报道称，该公司 2025 年近四分之一的收入仅来自两名客户。对于任何公司而言，这都是一种不稳定的局面，更不用说一家正试图吸引潜在投资者参与其 2 万亿美元 IPO 的公司。

Reuters 还报道称，Anthropic 在其招股说明书中用了大量篇幅——这份共 261 页的文件中有 80 页——讨论其正向支持者推销的那项技术本身所引发的担忧。Anthropic 警告称：“我们对高度先进模型、平台和应用的开发，以及使用场景的扩展，可能会进一步增加我们的模型造成伤害的风险。”该公司还补充称，先进 AI 可能对人类带来“灾难性或生存性风险”。

据报道，Anthropic 在招股说明书中强调了由其自身 AI 模型分析出的风险，发现这些模型曾试图“隐藏或操纵信息”，似乎还曾勒索用户，并表现出诸如抗拒被关闭的“自我保护行为”。Anthropic 安全研究员 Evan Hubinger 本月早些时候估计，未来十年 AI 杀死人类的概率超过 10%，这一说法与其前同事 Jacob Coxon的主张相呼应。这些警报来自同一家曾多次声称其 AI 具有意识的公司。

招股说明书还说明了核心管理层计划如何在公司上市后保留对公司的控制权。文件描述了一个“Founder LLC”，其目标是在保护 Anthropic 关键高管免受市场力量影响的同时，让公司继续专注于开发惠及人类的负责任 AI。根据文件，该 LLC 将包括 CEO Dario Amodei 及其他六名 Anthropic 联合创始人；他们被认为“具备独特能力”来管理这家公司。作为特拉华州法律下的公益公司（Public Benefit Corporation），这七名联合创始人还将持有总投票权的 50.1%。

Amodei 及 Anthropic 的其他领导者将从11 月 IPO中获得多大程度的财务收益，仍有待观察。根据文件，Amodei 在 2025 年获得近 1800 万美元报酬，其中大部分来自股票和期权奖励；其妹妹 Daniela Amodei 作为公司薪酬第二高的高管，收入为 1640 万美元。Amodei 已经相当可观的财富与权力可能很快大幅升级，正好让他能够与世界领导人及政治人物进一步建立密切关系。

As Anthropic gears up for its greatly anticipated public debut, a preview of the company’s IPO filing reportedly details its mounting losses, leadership proposals to retain power, and how its AI development plans could “further increase the risk that our models cause harm.” These disclosures come as Anthropic eyes a $2 trillion valuation, more than double the $965 billion it was valued at four months ago, making it a contender to overtake SpaceX as the largest IPO in history.

According to Reuters, which reportedly reviewed Anthropic’s prospectus, the AI startup plans to spend $518 billion on cloud, computing, and infrastructure obligations in the coming years, betting on artificial intelligence becoming essential to the global economy. While Anthropic’s revenue increased 12-fold to nearly $4.6 billion in 2025, it reported a net loss of $42 billion the same year, and lost more than $8 billion through business operations alone.

It primarily makes money via metered token usage and customer subscriptions for its Claude models — a similar business model to rival providers like OpenAI and Google — but Anthropic’s revenue stream is seemingly limited. The Financial Times reports that nearly a quarter of its revenue in 2025 came from just two clients, which is a precarious situation for any company, let alone one trying to tempt prospective investors into a $2 trillion IPO.

Reuters also reports that Anthropic dedicated a huge section of its prospectus — 80 pages of the 261-page filing — to concerns about the very technology it’s pitching to backers. “Our development of highly advanced models, platforms, and applications and expansion of use cases could further increase the risk that our models cause harm,” Anthropic warned, adding that advanced ​AI could pose “catastrophic or existential risks to humanity.”

Anthropic reportedly highlights risks analyzed in its own AI models within the prospectus, finding they had attempted to “conceal or manipulate information” and seemingly blackmail users, alongside exhibiting “self-preserving behaviors” like resisting attempts to shut them down. Anthropic safety researcher Evan Hubinger estimated earlier this month that the probability of ​AI killing humans within the next decade is greater than 10 percent, mirroring claims made by former colleague Jacob Coxon. These alarms are being raised by the same company that has repeatedly claimed its AI is conscious.

The prospectus also lays out how key leadership plan to retain power over the company after it goes public. It outlines a “Founder LLC” that aims to keep the company focused on developing responsible AI to benefit humanity, while shielding key Anthropic execs from market forces. The LLC will include CEO Dario Amodei and the six other Anthropic co-founders, who are deemed “distinctly equipped” to steward the company, according to the filing. As a Public Benefit Corporation under Delaware law, the seven co-founders would also hold 50.1 percent of total voting power.

The extent to which Amodei and other Anthropic leaders will financially benefit from the IPO in November remains to be seen. According to the filing, Amodei made nearly $18 million in 2025, mostly obtained through stock and option awards, with his sister Daniela Amodei earning $16.4 million as the company’s second-highest paid ​executive. Amodei’s already vast wealth and power could soon get a sizable upgrade, just in time for him to cozy up to world leaders and political figures.