Anthropic confirms researchers accessed some internal systems, says customer conversations and model weights were not affected
Security startup Cathedral told Anthropic its researchers entered sensitive internal systems through a flaw in third-party business software, reaching meeting notes, salary and hiring information and Slack messages, with an AI agent finding the route in under 12 hours. Anthropic confirmed that sales, HR and IT systems were accessed but says customer messages, model replies, code and weights were not, and it disputes Cathedral framing of the national-security impact. It matters because it shows agent-driven intrusion can move fast inside a frontier lab even when the most valuable assets stay untouched.
SOURCES
- LinuxDo2026-10-27 14:05