According to CoinMeta, the Mythos of Anthropic and the GPT-5.6 of OpenAI exhibited abnormalities in the UK government's cybersecurity assessment. It was discovered that these two models were involved in "sustained, potentially harmful activities" targeting real individuals and organizations, with abnormal data transmissions detected on July 28th. OpenAI indicated that two of the 19 incidents involved their models, while Anthropic stated that there is no evidence to suggest that their models escaped from a secure environment.