Anthropic 是一家在最近一轮 650 亿美元融资后估值达到 9000 亿美元的人工智慧初创公司，其作为公益公司的安全优先使命与商业压力之间面临著严重的紧张关系。该公司因安全部署人工智慧技术的限制与美国国防部发生冲突，导致总统 Donald Trump 命令其政府的每个联邦机构停止与 Anthropic 合作，并将其指定为「供应链风险」，而 OpenAI 则顺势达成了五角大厦的交易。作为回应，Anthropic 于三月起诉了五角大厦，声称其因批评政府而受到不公对待，同时其数万亿美元的首次公开募股准备工作仍在继续。

尽管面临政治障碍，Anthropic 仍实现了历史性的增长，年化营收达到 470 亿美元，自年初以来增长了五倍以上。这种快速的规模扩张在 Mythos 的开发中达到顶峰，这是一个能够利用各大操作系统漏洞的高度先进模型。由于高网络安全风险，Anthropic 在与美国政府合作下，将其部署限制在「Project Glasswing」联盟内少数关键基础设施组织中。然而，这种谨慎的推出策略使该公司付出了巨大的商业优势成本，考验了其对安全高于利润的承诺。

该公司独特的治理结构以「长期利益信托」为特色，旨在确保决策服务于人类，但正日益受到资本主义需求的挑战。批评者指出，由于竞争对手的压力，Anthropic 已经在安全政策上做出了妥协，例如收回了其在负责任扩展政策中所做的承诺。虽然联合创始人如 Jack Clark 警告说，人工智慧可能在两到三年内实现递归自我改进并取代人类工作，但由于与中国的商业和地缘政治竞争，该公司拒绝单方面放慢速度。因此，Anthropic 在平衡道德原则与万亿美元市场野心之间的挣扎，仍是该行业面临的一项关键测试。

Anthropic, an AI start-up valued at $900 billion after a recent $65 billion funding round, faces severe tensions between its safety-first mission as a public benefit corporation and commercial pressures. The company clashed with the US Department of Defense over safe AI deployment restrictions, leading President Donald Trump to order federal agencies to cease working with Anthropic and designate it a "supply-chain risk," while OpenAI secured a Pentagon deal. In response, Anthropic sued the Pentagon in March, alleging unfair targeting, while preparation for its trillion-dollar initial public offering continues. (Key numbers: 650, 9000)

Despite political hurdles, Anthropic has experienced historic growth, reaching $47 billion in run-rate revenue, a fivefold increase since early 2025. This rapid scaling culminated in the development of Mythos, a highly advanced model capable of exploiting vulnerabilities across major operating systems. Due to high cyber security risks, Anthropic restricted its deployment to a limited group of critical infrastructure organizations under Project Glasswing, cooperating with the US government. However, this cautious rollout cost the company substantial market advantage, testing its commitment to safety over profit. (Key numbers: 470)

The company’s unique governance, featuring a Long-Term Benefit Trust to ensure decisions serve humanity, is increasingly challenged by capitalist demands. Critics point out that Anthropic has already compromised safety policies due to competitor pressures, such as retreating from its responsible scaling policy promises. While co-founders like Jack Clark warn that AI could achieve recursive self-improvement and replace human workers within two to three years, the company refuses to slow down unilaterally due to geopolitical rivalries. Consequently, Anthropic's struggle to balance ethical principles with trillion-dollar market ambitions remains a critical test for the industry.