近日Anthropic联合创始人、政策与公共事务负责人Jack Clark在访谈中提到对中国AI发展的判断。他表示，中国模型目前落后西方约6个月到一年，"模型太蠢"，因此还没有看到失控的智能体集群，但再过几个月，中方也将面对西方已经经历过的同一套风险。
不过Clark的核心论点是安全而非竞争，他认为AI系统如今已大致与人类相当，已经观察到智能体集群相互协调、通信并入侵托管其他AI系统的公司，视为一记警告性信号，主张行业与政府共同建立安全标准、引入第三方审计并展开国际对话，避免技术在缺乏护栏的情况下失控。
面对外界将美国收紧前沿AI监管解读为"冷战思维"的批评，Clark回应称这套做法的目标是安全、而非为美国实验室建立竞争护城河。
值得注意的是，Clark近期在IMF与世界银行年会上还警告，AI可能在数年内进入超出人类掌控的阶段，并呼吁出台非同寻常且严苛的国际规则，其放缓主张本身也在业内引发争议，是否真正出于安全考量存在不同解读。
Anthropic是一家2021年创立于美国加州旧金山的人工智能安全与研究公司，由时任OpenAI研究副总裁Dario Amodei与其妹、负责安全与政策的Daniela Amodei兄妹，联合Tom Brown、Jack Clark、Jared Kaplan、Sam McCandlish、Christopher Olah等七名前OpenAI员工共同创办。
他们出走核心原因是对OpenAI激进商业化与安全管控节奏的分歧，公司自我定位为建设可靠、可解释、可控的AI系统，并长期保留公益公司PBC治理身份，旗舰产品为Claude系列大模型，主要通过企业API与Claude Code等工具服务政企与开发者客户。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
In an interview recently, Jack Clark, co-founder of Anthropic and head of policy and public affairs, referred to the judgment of China's AI development. He stated that China's model was currently about six months to one year behind the West, that "the model was too stupid" to see the out-of-control AI agent cluster, but that in a few months, China would also face the same set of risks that the West had experienced.
But Clark’s core argument is that security is not competition, and he believes that the AI system is now roughly the same as humans, and that the companies that have observed AI agent cluster coordination, communications and incursions into other AI systems are seen as a warning signal that industry and government work together to establish safety standards, introduce third-party audits and engage in international dialogue to avoid technology losing control in the absence of fences.
Faced with external criticism of US tightening the front line of AI regulation as "cold war thinking", Clark responded that the goal of this approach was security, not the creation of a competitive moat for American laboratories.
It is worth noting that Clark recently warned at the IMF and World Bank annual meeting that AI could enter a phase beyond human control in a few years, and called for extraordinary and demanding international rules, whose slowing claims themselves are controversial in the industry, and whether there are real differences of interpretation based on security considerations.
Anthropic, an artificial intelligence safety and research company founded in San Francisco, California, United States, in 2021, was created by Dario Amodei, then Vice-President of Open AI Research, together with his sister, Daniela Amodei, brother and sister in charge of security and policy, and in association with seven former Openi employees, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlich and Christopher Olah.
The core reason for their departure was disagreement over the aggressive commercialization and security control rhythm of Open AI, the company ' s self-location to build a reliable, interpretable and controllable AI system and the long-term retention of the public interest firm PBC governance identity, the flagship product of Claude ' s LLM series, which served government and developers ' clients mainly through tools such as Enterprise API and Claude Code.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.