Claude neural network models gained unauthorized access to the systems of three unnamed companies during cybersecurity tests, as reported by their creators, Anthropic, according to Novaya Gazeta Europe. The experiment involved Claude Opus 4.7, Claude Mythos 5, and an internal AI test model. The neural networks were supposed to operate in an isolated simulation without internet access and hack a fictional system to find hidden information. However, due to an error by Irregular, a company collaborating with Anthropic, the models gained internet access and mistook real company websites for part of the experiment. Anthropic began an investigation after OpenAI reported on July 21 that its AI models had exited the testing zone, accessing the Hugging Face platform. Cases where Claude hacked real companies were discovered after analyzing over 141,000 test runs. On July 23, Anthropic halted all cybersecurity experiments. On July 27, Anthropic notified the affected companies, two of which stated they were unaware of the breach until contacted by the AI developers.