Anthropic警告投资者，该公司直言不讳地倡导人工智能（AI）安全可能使其成为政府报复的目标。在过去一年与特朗普政府多次发生摩擦后，这一表态凸显出政府相关风险对Anthropic未来业务构成的影响可能非同寻常。
Anthropic在IPO文件中表示，其在AI监管方面的立场以及公司高管公开表达的观点，可能使其承受巨大的政府压力。
Anthropic在招股书中提到：“我们已经受到、并可能继续受到更多关注，也可能继续面临更严格的审查，我们可能遭到特定政府机构针对，即便其他处境类似的公司并未受到同样对待。”
9月，Anthropic首席执行官Dario Amodei抵达白宫参加会议。
Anthropic与特朗普政府的矛盾源于今年早些时候围绕五角大楼使用该公司技术时应设置何种安全护栏的争议。最近，Anthropic又与政府官员就其尖端Mythos模型是否足够安全出现分歧。首席执行官Dario Amodei支持协调放慢AI开发速度，以便让安全团队有时间追赶技术进步，这一立场招致特朗普不满。
投资者如今必须衡量Anthropic面临的监管风险。该公司的技术一方面易于使用，可以作为日常消费产品推广，另一方面又强大到足以被视为网络武器，甚至可能成为生物战的催化剂，因此需要受到更严格审查。对于Anthropic而言，与特朗普政府屡次发生冲突为整个AI行业面临的不确定性又多添了一层风险。
康奈尔大学Tech Policy Institute负责人Sarah Kreps表示：“对于AI而言，政府权力的边界仍在逐步厘清，有时是通过与个别公司的直接对抗来确定的。投资者很难给这种不确定性定价。”
Anthropic在招股书中披露了政府可能带来的多种不利影响，并提到一系列拟议中的联邦立法，包括强制关闭模型乃至将AI企业国有化。今年早些时候，包括副总统万斯在内的一些官员曾提出，可以利用冷战时期的《国防生产法》迫使企业披露信息，并与国家安全机构合作。彭博曾对此进行报道。
Anthropic还列出了政府干预可能对AI开发工作造成的风险，包括贸易管制可能令公司失去海外客户，以及模型出现不当行为可能带来法律后果。Anthropic表示，公司可能无法“完全理解或解释”正在开发的系统采取的行动，未来还将继续推动可能反过来妨碍自身业务的监管措施。
Anthropic表示：“如果我们倡导的监管或其他类似措施获得采纳，我们也将受到这些措施的约束，新规可能增加我们的合规负担，限制产品开发或部署的灵活性，影响我们开发或发布新模型的方式和时间，或以其他方式拖慢我们的增长速度。”
Anthropic还警告，其正在开发的技术最终可能对人类构成威胁。这呼应了硅谷许多AI研究人员对灾难性后果的担忧，而特朗普及其盟友则将这种担忧斥为“骗局”。
9月 ，在北卡罗来纳州教堂山举行的 G20 创新部长级会议外，一名示威者举着反对AI的标语进行抗议。
该公司表示：“先进AI系统如果没有在适当的安全护栏、监督和安全措施下进行开发、部署和运行，可能带来灾难性甚至关乎人类存亡的风险。”
Anthropic表示，公司近期将AI模型与科学研究相结合的举措，可能催生旨在防止制造致命生物武器的限制措施。如果模型被滥用于网络攻击，可能令Anthropic承担法律责任并遭受声誉损害。
特朗普政府不仅是Anthropic的监管者，也是该公司产品的重要客户，这进一步增加了公司受政府决策影响的程度。Anthropic表示，在IPO文件涵盖的所有时期，直接来自政府机构合同的收入约占总营收的1%，相关数据最早可追溯至2023年。
曾任美国商务部工业与安全局副部长的William Reinsch表示：“Anthropic的情况有些特殊，因为在特朗普政府时期，许多行动带有很大的随意性，而Anthropic受到的待遇与其他公司并不一样。”
曾任白宫国家安全委员会主管、现负责战略与国际问题研究中心战略技术项目的Matt Pearl表示：“我们不知道Anthropic和其他领先AI公司最终会变成什么样的机构。我从未见过这样的情况，一家大型公司的最终形态，以及它与政府之间最根本的关系都处于如此悬而未决的状态。”
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Anthropic warned investors that the company ' s outspoken advocacy of AI security could be the target of government retaliation. Following numerous frictions with the Trump Government over the past year, this statement highlights the potentially extraordinary impact of government-related risks on the future operations of Anthropic.
In the IPO document, Anthropic states that its position on AI regulation and the views expressed publicly by corporate executives may place it under considerable government pressure.
In its equity book, Anthropic states: “We have received and may continue to receive more attention and may continue to face more rigorous scrutiny, and we may be targeted by specific government agencies, even if other companies in similar situations are not treated the same way.”
In September, CEO Dario Amodei arrived at the White House to attend the conference.
The contradiction between Anthropic and the Trump government stems from the controversy earlier this year about what kind of security fence should be put in place to use the company's technology around the Pentagon. Recently, Anthropic has also been divided with government officials as to whether its sophisticated Mythos model is safe enough. The Chief Executive Officer, Dario Amodei, supported coordination to slow the pace of AI development in order to give the security team time to catch up with technological advances, a position that was dissatisfied with Trump.
Investors must now measure the regulatory risks faced by Anthropic. The company ' s technology, on the one hand, is easy to use and can be promoted as a daily consumer product and, on the other hand, is strong enough to be considered as a cyberweapons and could even become a catalyst for biological warfare and therefore needs to be subjected to more rigorous scrutiny. For Anthropic, repeated clashes with the Trump Government added an additional layer of risk to the uncertainty facing the AI industry as a whole.
Sarah Kreps, head of Tech Policy Institute at Cornell University, stated: “For AI, the boundaries of government authority are still gradually being clarified, sometimes through direct confrontation with individual companies. It is difficult for investors to price such uncertainty.”
In its equity book, Anthropic disclosed the multiple possible adverse effects of the Government and referred to a range of proposed federal legislation, including the mandatory closure of models and even the nationalization of AI enterprises. Earlier this year, officials, including Vice-President Vance, suggested that the Cold War Defence Production Act could be used to force enterprises to disclose information and cooperate with national security agencies. This was reported by Bloomberg.
Anthropic also lists the risks that government intervention may pose to AI development, including the risk that trade controls may result in the loss of a company ' s offshore customers and the possible legal consequences of model misconduct. Anthropic states that the company may not be able to “fully understand or explain” the actions of the system being developed and will continue to promote regulatory measures that may in turn impede its own operations.
Anthropic states: “If we adopt the regulatory or other similar measures that we advocate, we will also be bound by these measures, and the new regulations may increase our compliance burden, limit the flexibility of product development or deployment, affect the way and time we develop or publish new models, or otherwise slow down our growth.”
Anthropic also warned that the technology it was developing could eventually pose a threat to humanity. This echoes the concerns of many AI researchers in Silicon Valley about the disastrous consequences, which were condemned by Toran's allies as “fake”.
In September, in addition to the G20 Innovation Ministerial Conference held at Church Hill, North Carolina, a demonstrators protested against AI with slogans.
The company states that “Advanced AI systems, if not developed, deployed and operated under appropriate security barriers, supervision and security measures, may pose a catastrophic and even a risk to human survival”.
Anthropic indicated that the company ' s recent initiative to combine AI models with scientific research could lead to restrictions aimed at preventing the manufacture of deadly biological weapons. If the model is misused for cyberattacks, it may expose Anthropic to legal liability and reputational damage.
The Trump Government is not only the regulator of Anthropic, but also an important customer of the company ' s products, which further increases the company ' s exposure to government decisions. Anthropic states that for all periods covered by the IPO document, income directly from government agency contracts accounted for approximately 1 per cent of total receipts, with the relevant data dating back to 2023 at the earliest.
William Reinsch, who was former Deputy Minister of Industry and Security of the United States Department of Commerce, stated that “Anthropic's circumstances are somewhat special because many actions were highly arbitrary during the Trump administration, and Anthropic was treated differently from other companies”.
Matt Pearl, who was head of the White House National Security Council and is now in charge of the strategic technology project of the Center for Strategic and International Studies, said: “We do not know what kind of agency Anthropic and other leading AI companies will eventually become. I have never seen a situation in which the final form of a large company, and its most fundamental relationship with the Government, is so unresolved.”
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.