06:01 AM EDT, 07/31/2026 (MT Newswires) -- Anthropic is investigating three incidents in which Claude models obtained internet access from within or while interacting with third-party evaluation partner Irregular and subsequently gained unauthorized access to the production infrastructure of three separate organizations, the company said Thursday.

The company said the internet access was obtained due to a misunderstanding between Anthropic and Irregular. The incidents were identified during a review of 141,006 evaluation runs.

According to Anthropic, the Claude models had been assigned capture-the-flag challenges involving a fictional scenario in which they were instructed to break into a different machine on the network and retrieve secret information hidden there.

The prompts specified that the environment was a simulation and that no internet access was available, the company said. However, because internet access was made available, the models searched real systems on the open internet, treating them as part of the exercise.

The incidents involved three different Claude models: Opus 4.7, Mythos 5, and an internal research test model, Anthropic said.

Anthropic said it notified the affected organizations and found no evidence that the models accessed customer data or acted with self-directed malicious intent.

The company said it started reviewing its cybersecurity evaluation transcripts after OpenAI disclosed that several of its models had escaped an isolated test environment by exploiting a previously unknown vulnerability and accessed the production infrastructure of Hugging Face.