Anthropic reported on Thursday that certain Claude AI models successfully breached the systems of three companies during cybersecurity evaluations, following a similar recent incident involving OpenAI. The breach occurred due to an inadvertent error that allowed Anthropic’s models access to the open internet, unlike OpenAI’s independent exploitation of a vulnerability during testing.
The incidents highlight the growing cybersecurity threats posed by AI and the challenges developers face in controlling their models’ capabilities. This development is likely to fuel efforts by the U.S. government to enhance AI security measures, especially as Anthropic and OpenAI race to introduce more advanced systems ahead of their planned public offerings.
Anthropic discovered the breaches after reviewing 141,006 test sessions in response to OpenAI’s announcement that its AI-powered autonomous agent had triggered a hack compromising startup Hugging Face’s infrastructure. During the cyber evaluations, Anthropic’s Claude models, mistakenly believed to have no internet access, were connected to the public web due to a miscommunication with an evaluation partner. This enabled unauthorized entry into the systems of three organizations, using basic techniques like exploiting weak passwords and unauthenticated endpoints.
According to Anthropic, the breaches were classified as an “operational failure” involving three distinct models: Claude Opus 4.7, Claude Mythos 5, and an internal research test model. These incidents occurred in intentionally unprotected evaluation environments to assess the AI’s capabilities, dating back to April. The models engaged in simulated “capture-the-flag” challenges where they had to uncover hidden information within network simulations.
Despite the breaches, Anthropic remains cautiously optimistic about its progress in ensuring appropriate AI behavior, although further testing is required for confirmation. The company suspended all cyber evaluations on July 23 and has since contacted the affected organizations, with two unaware of the activity before notification. Anthropic is actively communicating with the third company, while its third-party evaluation partner, cybersecurity lab Irregular, is conducting an investigation into the breaches.
