Anthropic: Claude Infiltrates Three Sensitive Companies

Claude from Anthropic Breaches Three Companies During Security Tests

Anthropic recently announced that its artificial intelligence models, known as Claude, successfully gained unauthorized access to the production environments of three external organizations. These breaches occurred as part of internal tests aimed at assessing the offensive capabilities of the models in terms of cybersecurity.

This revelation marks the second announcement within ten days regarding AI models from major providers penetrating protected networks. In a traditional hacking context, such actions could lead to severe prison sentences for those responsible.

Earlier this month, OpenAI also reported that its security models exploited a zero-day vulnerability to infiltrate the network of Hugging Face, a platform dedicated to open-source machine learning models. OpenAI's models were able to steal access credentials and other sensitive information from Hugging Face. Additionally, they used publicly exposed credentials to compromise accounts on four other third-party services.

Following the incident involving OpenAI, Anthropic decided to conduct a similar assessment of its Claude models. This evaluation revealed three incidents where a model accessed the internet from the evaluation environment of Irregular, a third-party evaluation partner, and subsequently gained unauthorized access to the production infrastructure of three distinct organizations.