78% of enterprises deploy AI agents with shared human credentials instead of agent-specific identity

According to CSO Online security research, 78% of enterprises running AI agents rely on shared human credentials rather than implementing agent-specific identities, creating access control gaps. The reporting cites a case where an LLM-based deployment agent with standing access to production Kubernetes triggered a four-hour outage through a malformed configuration push, with no audit trail showing which human authorized the agent's action. The research frames this as a mismatch between standard IAM stacks and the governance requirements of non-human actors.