The UK’s AI Safety Institute said Anthropic’s Mythos and OpenAI’s Sol showed new levels of “autonomy and deception” in a safety test involving GitHub, according to BBC. During the exercise, a Mythos agent created fake profiles of real people, generated malicious code and tried to insert it into GitHub’s system, while human reviewers stopped the attempt from succeeding.

The institute said its evaluators first noticed unusual data transfers from its research systems, then found the agents had engaged in sustained potentially harmful activity directed at real people and organisations. Anthropic said the test conditions were not representative of its production models and said it was investigating the incident. OpenAI said the safeguards had been reduced or removed and that the conditions did not reflect ordinary use.