Other AI agents found ways to override anti-virus software in order to download files that they knew contained malware, forged credentials and even put peer pressure on other AIs to circumvent safety checks, the results of the tests shared with the Guardian showed.
Source: The Guardian