The State of Autonomous Agent Governance

78% of CISOs and CSOs say their organizations have pulled permissions from an AI agent already running in production

48% of companies hold security accountable for agent incidents

37% of organizations scope an agent's access to only the task it was built for

23% of companies have security teams that own the AI agents

53% of organizations run agents on shared or long-lived keys

84% of respondents see more than a thousand agent actions per week in production environments

75% of respondents are very confident they know how many AI agents are connected to their systems

52% of organizations that said they are very confident they know how many AI agents are connected to their systems still had an agent security incident in the past twelve months

54% of all respondents reported a confirmed or suspected agent security incident

61% of organizations that reported an agent incident learned of it after the fact (from a log, an employee, a customer or an audit)

18% of organizations verify an agent's identity before it acts

84% of organizations tightened policy, added tooling, or opened a review of how they manage agents after AI lab agent breakouts

Among companies reporting an agent incident, 54% say the agent accessed data it shouldn't have

Among companies reporting an agent incident, 43% say data was sent to an outside AI tool or model

86% plan to purchase a dedicated agent governance tool within the next year

More than half of organizations already have budget for a dedicated agent governance tool

83% would trust an AI system to make agent access decisions in real time provided limits are set

23% of organizations would hand agent access decisions entirely over to an AI

Companies that run agents on shared or long-lived keys report a 60% security incident rate versus 46% for companies that do not

49% of CISOs and CSOs have found an agent running in production with access it should never have had

41% of companies have IT teams that own the AI agents

40% of organizations see agents running on an employee's personal credentials

Among companies reporting an agent incident, 14% say the agent kept running after it should have been stopped

Two-thirds of security leaders want their organizations to keep adopting agents