Anthropic Confirms Internal Systems Accessed via Third-Party Vulnerability
Cybersecurity startup Cathedral used AI agents to exploit a third-party software vulnerability, gaining access to some of Anthropic's internal HR, sales, and IT systems within 12 hours. This incident highlights the growing security risks associated with third-party software supply chains and the increasing capability of AI agents to autonomously discover and exploit vulnerabilities. Anthropic stated that core assets such as customer conversations, model weights, and source code remained completely secure, and the affected third-party software has been disabled pending a patch.
## BACKGROUND
Model weights are the learned numerical parameters within an artificial neural network that determine how inputs are converted into predictions, forming the core intellectual property of advanced AI models. AI agents are autonomous software entities powered by large language models capable of planning, using tools, and executing complex workflows to achieve specific goals.