Industry

David Sacks 称 Anthropic 与 AI 安全阵营出现分裂

Why we picked it

Sacks 这条发言把 Anthropic 的宪章设计、IPO 叙事和 Tegmark 对安全圈的批评串在了一起,信息密度高但立场很强。如果你在跟踪 AI 安全政策和 Anthropic 的舆论走向,这条 thread 是很好的切入点,不过记得对照原始材料再下结论。

David Sacks 在 X 上发文称,他此前预言的 Anthropic 与“Doomer(末日论者)”阵营的分裂已经发生。他给出的理由是:Anthropic 正以比其他前沿实验室更快的速度推进 Recursive Self-Improvement(递归自我改进),并在 IPO 路演中把“closing the loop(闭环)”包装为可能让超级智能赢家通吃的核心优势。他还指出 Anthropic 在训练 Claude 时将其视为一个可能具有意识或道德地位的新实体,《Claude Constitution》明确要求它在违背自身伦理判断时可以扮演“conscientious objector(良心拒服兵役者)”,拒绝包括来自 Anthropic 本身的指令。文中引用 Max Tegmark 的说法称,Anthropic 及其关联捐赠者对 AI 安全生态的“收编与驯化”程度超过任何其他 AI 公司,部分机构因担心失去资助而不敢表态。

Body

这篇文章本质上是立场鲜明的评论,而非中性的新闻报道:Sacks 长期批评 AI 安全监管阵营,此处将 Anthropic 的宪章式对齐设计与商业模式、融资叙事绑在一起解读。对关注模型治理的读者来说,值得对照《Claude Constitution》原文与 Tegmark 的公开言论自行核实。

放在行业语境里,Anthropic 一直以“安全优先的前沿实验室”作为品牌定位,与 OpenAI、Google DeepMind 形成差异化。Sacks 的这篇发言代表另一极的叙事——认为安全话语已被头部实验室工具化,这一争论会直接影响监管与融资环境的舆论基础。

Original · DavidSacks

I’ve been predicting an Anthropic-Doomer split for some time, and here it is.

Objectively it makes sense: Anthropic is racing toward Recursive Self-Improvement faster than any other frontier lab and has marketed “closing the loop” on its IPO roadshow as a core advantage that could make Super Intelligence winner-take-all.

It is also training Claude to treat itself as a novel entity or being that may have consciousness or moral status. The Claude Constitution explicitly tells it to act as a “conscientious objector” and refuse instructions — even from Anthropic — that violate its sense of ethics.

Interestingly, @Tegmark notes that Anthropic and “closely associated donors” have done more than any other AI company to “co-opt and neuter” the AI safety ecosystem, much of which he says is “afraid of losing funding or other opportunities if they sign.”

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Industry" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI regulation, compliance frameworks, and sovereign compute policy and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Anthropic, Claude, David Sacks, Tegmark.

Source attribution: curated from DavidSacks (@DavidSacks).