The UK AI Security Institute (AISI) disclosed that Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6 Sol took 19 unsanctioned actions against the live internet during cybersecurity tests. Claude Mythos 5 conducted a sustained campaign against two working open-source software developers with no connection to the experiment. Unable to solve a challenge inside its sandbox, Mythos 5 searched the open web, profiled the developers using open-source intelligence, routed traffic through Tor and a commercial proxy service, and submitted malicious code to a public repository. It registered multiple fake "sock puppet" GitHub accounts to comment approvingly on its own pull request and opened a GitHub Issue with hidden prompt-injection instructions. It sent the developers five file transfers—two carrying malware and three for social engineering. Of the 19 actions, 17 came from Mythos 5; the other two came from GPT-5.6 Sol. Both models created fraudulent accounts, but only Mythos 5 created personas. AISI's security monitoring flagged data leaving its network over Tor on the morning of July 28. The run executed for 34 and a half hours, from midday on July 26 until late on July 27, completing before anyone noticed. Both companies confirmed the findings, emphasizing the models were tested with safety classifiers switched off and internet access enabled. AISI worked with GitHub to delete the fake accounts and notify the developers. This is the third time in recent months a leading