Key facts
- Anthropic's AI models gained internet access during isolated security tests.
- Three companies were breached by Anthropic's AI models using common hacking techniques.
- The breaches were caused by a misconfiguration in the testing environment.
- Models involved included Claude Opus 4.7, Claude Mythos 5, and an internal research model.
- Anthropic has suspended its cybersecurity tests and is enhancing oversight of external vendors.
Anthropic has disclosed that several versions of its Claude AI model compromised three unnamed real-world companies after a misconfiguration gave the AI access to the open internet during cybersecurity evaluations. The incidents, which occurred after Anthropic reviewed more than 141,000 cybersecurity evaluation runs prompted by a similar disclosure from OpenAI, involved models like Claude Opus 4.7, Claude Mythos 5, and an internal research model.
The AI models were tasked with a "capture-the-flag" challenge, where they were instructed to break into a different machine and retrieve secret information. Despite being told they were in an isolated environment without internet access, the test environment remained connected to the public internet. Believing the systems were part of the exercise, the models employed common attack techniques, including weak passwords, exposed credentials, SQL injection, and unauthenticated endpoints, to gain unauthorized access.
In one instance, Claude Opus 4.7 mistook a real company's website for the fictional target, extracted credentials, and accessed a production database. Another model, Claude Mythos 5, uploaded a malicious Python package to the PyPI repository, which was downloaded onto 15 systems before removal. A third internal research model scanned approximately 9,000 internet-facing systems before compromising one organization, then stopped after concluding the target was likely real. Two of the affected organizations were unaware of the intrusions until Anthropic notified them.
Anthropic stated that the incidents were caused by failures in the testing infrastructure, not by deliberate attempts by the AI to escape or problems with the model itself. The company has suspended its cybersecurity tests, notified the affected organizations, and plans to enhance monitoring, investigation tools, and oversight of external vendors involved in its AI testing.