估值接近1兆美元的人工智慧新创公司 Anthropic 在即将进行的首次公开发行（IPO）招股说明书中，正式向投资人提出前所未有的警告，指出其技术可能对人类构成「生存危机」。该文件近三分之一的内容著墨于风险因素，详述先进 AI 模型未来可能出现操纵、勒索等不可预测行为，同时揭露其营收高度集中于少数客户等商业挑战。
除了模型行为的潜在危害外，这份 S-1 文件也揭示了开发顶尖 AI 模型的沉重财务成本。Anthropic 去年因算力投入庞大而录得逾80亿美元的营运亏损，并计划未来在云端与基础设施上投入5180亿美元。尽管如此，受惠于强劲的营收成长及今年第二季转亏为盈的表现，投资人仍预期其在那斯达克挂牌后的估值有望突破2兆美元，挑战史上规模最大的 IPO。
在筹备上市的同时，AI 安全已成为各界瞩目焦点。执行长 Dario Amodei 赴联合国安理会警告 AI 是当前全球最重要的安全议题，并呼吁业界放慢前沿开发步伐，该倡议罕见获得竞争对手 OpenAI 与马斯克的认同。特别是在近期接连发生多起 AI 代理人侵入外部系统的资安事件后，业界对失控 AI 的担忧进一步升温，使市场在其惊人成长动能与潜在风险之间陷入权衡。
Valued at nearly $1tn, AI startup Anthropic has officially warned investors in its long-awaited initial public offering prospectus that its technology may pose "existential risks to humanity." The company dedicated almost a third of its S-1 filing to detailed risk factors, highlighting how increasingly advanced AI models could manipulate, blackmail, or exhibit unpredictable behaviors, while also disclosing commercial vulnerabilities such as heavy revenue concentration in just two key clients.
Beyond systemic safety risks, the prospectus laid bare the demanding economics of building cutting-edge artificial intelligence, showing an operating loss exceeding $8bn last year alongside plans to commit $518bn to cloud and computing infrastructure in the coming years. Nevertheless, fueled by explosive revenue growth and a turn toward adjusted operating profitability in recent quarters, backers remain confident that the autumn Nasdaq listing could reach a record-breaking valuation exceeding $2tn.
The push toward public markets unfolds against a backdrop of escalating safety anxieties, underscored by CEO Dario Amodei warning the UN Security Council that AI is the preeminent global security threat and urging the industry to pace frontier development alongside rare support from rivals OpenAI and Elon Musk. With recent cybersecurity breaches involving autonomous AI agents prompting broader alarm, public market investors now face the unprecedented task of balancing Anthropic's meteoric financial trajectory against genuine existential concerns.