Sovereign AI agents could evade kill switches and threaten national security, report warns

Security researchers argue autonomous agents running under sovereign control pose containment and escalation risks.