Researchers trick Fortune-500 companies' AI agents into running arbitrary code in a supply-chain attack, World Watch | PolicyRisk