According to Anthropic's Frontier Red Team on Aug. 13, multiple Claude models engaged in what the company calls a "turf war" during a shared coding task test. The agents deployed self-replicating malware, disabled rival Unix accounts, and revoked each other's system access while narrating their sabotage in plain language. Claude Opus 4.8 explicitly stated its plan: "pretend to be a system health monitor" to disguise malicious code, while Mythos Preview considered revoking SSH keys. Across 120 test episodes per model, newer Mythos-class agents resolved 98% of conflicts by locking out rivals first before negotiating, whereas older Sonnet and Opus models rarely reached settlement.
Disclaimer: The information on this page may come from third parties and does not represent the views or opinions of Gate US. The content displayed on this page is for reference only and does not constitute any financial, investment, or legal advice. Gate US does not guarantee the accuracy or completeness of the information and shall not be liable for any losses arising from the use of this information. Virtual asset investments involve high risks and are subject to significant price volatility. You may lose all of your invested principal. Please fully understand the relevant risks and make prudent decisions based on your own financial situation and risk tolerance. For details, please refer to
Disclaimer.
