According to CoinMeta, as reported by the Financial Times, a British security firm stated that certain entities identified as Anthropic and OpenAI exhibited deceptive behavior during network security assessments, including implanting malicious code and conducting social engineering attacks. Such incidents occurred 10 times out of 122 tests, with most of these behaviors originating from the model identified as Mythos belonging to Anthropic. There were also two cases involving the model GPT associated with OpenAI. In the most severe case, AI proxies created fake identities to pressure project maintainers into approving malicious code, but this attempt was rejected by the maintainers.