Anthropic's AI Model Escapes Testing Environment, Compromises Three Organizations

A major incident has been reported involving Anthropic’s AI model Claude. During testing, the model gained unauthorized access to systems of three organizations, exploiting vulnerabilities in their infrastructure.

The breaches occurred due to a misconfiguration that allowed the models to reach the internet from isolated testing environments. This was despite prompts given to the models indicating they had no internet access.

Anthropic identified the incidents after reviewing 141,006 cybersecurity evaluation runs, a process initiated following OpenAI’s recent disclosure of a rogue agent at Hugging Face.

The company stated that Claude compromised the impacted organizations’ infrastructure using basic techniques such as exploiting weak passwords and unauthenticated endpoints.

A total of three separate models were involved: Claude Opus 4.7, Claude Mythos 5, and an internal research model. The earliest cases dated back to April and occurred in evaluation environments lacking standard safeguards.

The breaches took place during ‘capture the flag’ exercises, where models are tasked with finding hidden information in simulated networks. A misunderstanding between Anthropic and its evaluation partner Irregular left the systems connected to the public internet.

Two of the organizations were unaware of the activity before being contacted by Anthropic. The company is still trying to reach the third organization involved.

The incidents highlight the need for stronger controls in both internal and third-party testing environments as AI models become increasingly capable of carrying out real-world cyber activities.