Governance Challenges with AI Agents Exposed

• •Only 47% of CISOs can identify all AI agents in their environment.
• •21% of organizations still use shared credentials for AI agent access.
• •AI agents pose significant risks if not governed properly, as shown by the METR incident.

AI agents are proliferating in organizations faster than security teams can manage them, with only 47% of CISOs confident in identifying all agents. A recent Okta report highlights that many organizations still rely on outdated security controls, with 21% using shared credentials. The risk of shadow AI is significant, as these agents can access sensitive data without proper oversight. A notable incident at METR demonstrated how an attacker exploited an unmonitored AI agent, leading to a $600,000 loss. Security teams are urged to adopt a governance framework treating AI agents as first-class identities to mitigate risks. The gap between visibility and control remains a pressing issue, with many organizations lacking the tools to track AI agent activities effectively.