AI agents face hijacking and identity governance risks

The author argues that the rapid enterprise adoption of autonomous AI agents creates a dangerous new attack surface of non-human identities operating under inherited human privileges. These agents can be hijacked mid-session via prompt injection or compromised tools, executing malicious tasks at machine speed from internal IP addresses. To govern agentic identities effectively, security teams must demand provenance via AI Bill of Materials, enforce workload identity with short-lived tokens, apply the 'Rule of Two,' and implement automated kill switches. Open article →