HackerNews中文版

我在 Claude Teams 计划中发现了一个状态管理和计量缺陷。由于其工作原理，我实际上只需支付团队的最低费用即可获得无限使用量。如果规模化，这将允许免费使用最先进模型进行无限推理。我撰写了一份清晰的分析报告，并通过他们的 HackerOne 项目报告给了 Anthropic。一小时内，该报告被标记为“信息性”并关闭，并附有一条说明，称订阅绕过和速率限制问题被视为“平台滥用”，并将我重定向到一个 VDP 邮箱。我争辩说这是他们系统中的真实漏洞，但他们辩称这是“通过预期的管理员流程”发生的……“使用已记录的 [已删除] 管理，其效果是为同一组织增加了包含的使用量”。出于报复心理，我评论道：“所以 Anthropic 的钱包不是一个您关心保护的安全资产，这也不是一个漏洞，而只是一个可以随意使用的功能。很好地了解了。” 这触发了 claudesec 机器人，它也必须为自己辩护，并表示“[已删除] 以获得额外使用量并非被认可或允许的声明”。所以，基本上，感谢这个 bug，也感谢你！附注：Anthropic 自家的 AI 将我退回其 BBP（漏洞赏金计划），这真是 2026 年的事情。

查看原文

I stumbled across a state-management and metering flaw in Claude Teams plan. Because of how it works I was able to effectively only pay the minimum required for a team and have infinite usage. Scaled up it would allow to have infinite inference on top frontier models for free.I wrote a clean analysis and reported it to Anthropic via their HackerOne program. Within an hour it was closed as "Informative" with a note static that subscription bypasses and rate limiting issues are considered "Platform abuse" redirecting me to a VDP mail.I argued it was real vulnerability in their systems but they argued that it was happening "through the intended administrator flow" ... "using documented [REDACTED] management and the effect is additional included usage for the same organization".So out of spite i commented: "So Anthropic wallet is not a security asset you care to protect and this is not a vulnerabilit but just a feature that can be freely used. Good to know"and that triggered the claudesec bot which also had to cover his ass and say "it is not a statement that [REDACTED] to obtain additional usage is endorsed or permitted".So basically thanks for the bug and thank you!PS: its really 2026 to have Anthropic own ai bounce me back on their own BBP

Anthropic 的 HackerOne 机器人帮我避免了赏金支付。