Anthropic reports leak of Claude account information at over 100 Korean companies

Anthropic reported to South Korea's Personal Information Protection Commission that account information for more than 100 Korean enterprise customers of its Claude generative AI service was exposed, the commission said on November 9; the filing was made on October 31 and remains in a preliminary review, with no formal investigation opened yet. According to the Washington Post, U.S. security startup Cathedral accessed Anthropic's internal systems through a vulnerability in third-party office software. Anthropic researcher Lydia Hallie said on X that the unauthorized access covered account data such as names, email addresses and subscription plans, but not prompts, model outputs or payment information, and that the company blocked the affected systems and patched the flaw the same day it learned of the incident. The disclosure matters because it tests enterprise confidence in Claude amid rising scrutiny of AI vendors' third-party security practices. What to watch next is whether the Korean regulator escalates to a formal probe and whether Anthropic details the scope of affected accounts and any remediation for enterprise clients.