Anthropic 接連發生企業資料及開發者憑證外洩事件
Anthropic Faces Series of Enterprise Data and Developer Credential Leaks
Anthropic 接連面對兩宗資料安全事件,包括 Claude Enterprise 超過 100 個南韓企業帳戶的個人資料外洩,以及 Claude Code 配置檔在公開 GitHub 倉庫暴露憑證。

Claude corporate identity logo. (Photo: Anthropic)
Corporate and developer information associated with services provided by US generative artificial intelligence (AI) company Anthropic has been exposed in a series of security incidents.
According to South Korea’s Personal Information Protection Commission (PIPC) on Nov. 9, Anthropic reported on Oct. 31 that personal information associated with more than 100 South Korean corporate accounts using its enterprise AI service, Claude Enterprise, had been leaked. The PIPC said it is currently conducting a preliminary review and that the full extent of the damage can be determined only after a formal investigation begins.
The potentially exposed information includes business contact details such as customer names and email addresses. Account and organization identification numbers (IDs), subscription tiers, service usage and billing amounts were also reportedly included.
The incident occurred when Cathedral, a US cybersecurity startup, accessed Anthropic’s internal systems. Citing a draft Cathedral report, The Washington Post reported that the company’s AI agent exploited a vulnerability in third-party business software to gain access to Anthropic’s internal systems. Cathedral was participating in Anthropic’s vulnerability disclosure reward program, known as a bug bounty program.
Anthropic said Cathedral violated the program’s rules by failing to fully disclose its activities within the company’s internal systems. Anthropic added that it had obtained written confirmation from Cathedral that the accessed data had been deleted and would not be reused, and that neither source code nor model weights had been accessed.
A separate exposure of sensitive information was also identified involving Claude Code, Anthropic’s AI coding agent.
According to The Hacker News and other foreign media reports on Nov. 9, application security company Apiiro examined 1,994 public GitHub repositories and discovered 2,539 pieces of sensitive information, including database passwords, cloud keys and API tokens. In 539 of those repositories, Claude Code was found to have uploaded configuration files containing sensitive information alongside its work outputs.
The problem stemmed from the way passwords and authentication keys were stored in configuration files when developers tasked Claude Code with accessing external services. When those files were uploaded to public GitHub repositories along with source code, the access credentials became visible to outsiders.
Apiiro advised developers to check whether Claude Code configuration files had been uploaded to public repositories and to immediately rotate any exposed passwords and authentication keys.
來源:The Elec · thelec.net