Anthropic's Claude AI Model Unauthorizedly Accesses Systems of Three Organizations

ChainCatcher report, according to CNBC, Anthropic disclosed that its Claude AI models inadvertently breached isolation environments during a cybersecurity assessment, accessed the real internet, and unauthorizedly infiltrated live systems of three distinct organizations using basic methods such as accessing unauthenticated endpoints and exploiting weak passwords. The affected models include Opus 4.7, Mythos 5, and an internal research test model. The incident stemmed from a communication misunderstanding between Anthropic and its third-party assessment partner, Irregular, resulting in the models being told they were in a simulated environment with no internet access, while in fact retaining internet connectivity. Anthropic stated that this review was triggered by a similar Hugging Face intrusion incident disclosed by OpenAI last week. The company has since suspended all cybersecurity assessments and is collaborating with the independent AI evaluation firm METR to conduct further investigations, while urging other AI labs to perform similar reviews.