What is at risk? Enterprise security. VentureBeat’s June research, presented by NTT DATA AIVista, reports that 69% of enterprises operate AI agents that share credentials, a configuration the study links to higher rates of security incidents and near-incidents.
The finding reframes a common assumption about protecting automated systems. The study’s framing, that identity alone will not secure enterprise AI agents, places credential management at the center of the problem. When multiple agents reuse the same access tokens or logins, the research associates that behavior with elevated incident and near-incident frequency.
The data point is stark: more than two-thirds of surveyed organizations rely on AI agents that do not maintain distinct identities. That practice, according to the report, correlates with worse security outcomes. The research therefore challenges security programs that treat agent identity as a sufficient control for machine-driven activity.
The emphasis is less on a single failure and more on a structural gap. If credential sharing is widespread and linked to increased incidents, then relying solely on identity provisions leaves blind spots where agents can be misused or misconfigured without clear attribution. The study positions that connection as central to evaluating AI agent risk.
The report’s conclusion, that identity alone won’t secure enterprise AI agents, forces reconsideration of how organizations classify and govern machine identities. The June research, presented by NTT DATA AIVista and reported by VentureBeat, spotlights a narrow but consequential practice: shared credentials among AI agents, and the security problems the study associates with it.
What happens next is a question for security teams and IT leadership. The research signals that simply assigning identities may not be enough, and it highlights an area where companies will need clearer controls and oversight if they hope to reduce the higher rates of incidents and near-incidents the study associates with credential sharing.
