Davos discussion mulls how to keep AI agents from running wild

Industry leaders at Davos discussed how to secure AI agents, highlighting risks such as prompt injection and insider-threat-like behavior, and advocating zero trust, least-privilege access, guardrails, and dedicated “guard agents” to monitor agent activity. Speakers emphasized leveraging extensive data signals and updated data infrastructure for proactive threat detection, noting that AI will be used by both attackers and defenders and that market activity is accelerating around AI-focused security solutions.