Anthropic Confirms Researchers Accessed Some Internal Systems, Asserts Customer Conversations and Model Weights Were Unaffected
Anthropic confirmed that some sales, HR, and IT systems were accessed by researchers; however, customer conversations, model responses, AI system code, and model weights were not affected. Cathedral claimed it gained access through a vulnerability in third-party commercial software, while Anthropic stated that Cathedral exaggerated the national security impact and did not fully disclose its activities.
Anthropic Confirms Researchers Accessed Some Internal Systems, Asserts Customer Conversations and Model Weights Were Unaffected
Cathedral, a cybersecurity startup founded by former DOGE members, reported to Anthropic that its researchers gained access to sensitive internal systems by exploiting a vulnerability in third-party commercial software. The AI agent found a pathway in less than 12 hours, reaching meeting notes, salary and hiring information, and Slack messages.
Anthropic confirmed that some sales, HR, and IT systems were accessed, but that customer messages sent to the model, the model's responses, the AI system's code, and the model's weights were not accessed. The company stated that its monitoring systems had detected some activity before Cathedral's report and promptly closed the access pathway. Anthropic believes Cathedral exaggerated the national security implications and did not fully disclose its activities within the systems. Cathedral, in turn, claims it followed standard security research practices and limited data access to the absolute minimum necessary.
Source: The Washington Post (https://www.washingtonpost.com/technology/2026/10/26/researchers-cybersecurity-start-up-cathedral-got-inside-anthropic/)