A Rubrik survey of 1,625 IT and security leaders found that 86% expect AI-agent adoption to outpace their organisations’ security guardrails within a year, with visibility, access controls and recovery capability emerging as key concerns.
Rubrik said its global survey of IT and security leaders found that 86% expect the use of AI agents to outpace their organisation’s security guardrails within the next year.
The finding comes from The State of the Agent: Understanding Adoption, Risk, and Mitigation, a report based on responses from 1,625 IT and security leaders. Rubrik’s research points to a growing concern among organisations deploying AI systems that can access data, interact with software and initiate business workflows.
AI agents differ from simpler generative AI tools because they may be assigned permissions to retrieve information, update records or trigger actions in connected systems. That autonomy can create operational benefits, but it also raises the stakes when an agent is misconfigured, given excessive access or influenced by an incorrect instruction.
Rubrik’s report frames the challenge as broader than protecting an AI model or application. Organisations need to know which agents are operating, which business systems they can reach, which identities they use and what actions they are authorised to perform.
Those questions are central to identity and access management. An agent connected to customer, financial or operational systems could have significant reach if it uses a highly privileged account. In that situation, an error or compromise could affect more than a single application or dataset.
IT Brief UK, reporting on the survey, also highlighted concerns about visibility into deployed agents and the ability to reverse their actions. The latter issue is particularly relevant for agents that can alter records, modify configurations or launch automated workflows across multiple systems.
The survey results suggest that governance for AI agents needs to include operational safeguards as well as decisions about whether a particular tool should be approved. For companies deploying such systems, that can mean maintaining an inventory of agents, documenting their data access and permissions, and monitoring their activity.
Applying least-privilege principles can help limit the potential effect of an unwanted action. Organisations may also need clear procedures for suspending agent access, investigating changes and restoring systems or records when automation produces unintended outcomes.
Recovery capabilities are especially important when agents are connected to sensitive business data or critical processes. Prevention controls can reduce risk, but they may not eliminate the possibility of mistaken instructions, faulty integrations or inappropriate access.
The research does not show that AI agents have already exceeded security controls at most companies. Instead, it records the expectations of respondents surveyed by Rubrik. As a vendor-sponsored study, the findings should be treated as an indicator of sentiment among the participating IT and security leaders rather than a definitive measure of the wider market.
Even so, the result highlights a practical issue for organisations moving from AI experimentation to production deployments. Security planning may need to progress alongside implementation, with agent identities, permissions, activity monitoring and recovery processes treated as core requirements rather than later additions.
Security teams expect guardrails to lag Rubrik said its global survey of IT and security leaders found that 86% expect the use of AI agents to outpace their organisation’s security guardrails within the next year.
The finding comes from The State of the Agent: Understanding Adoption, Risk, and Mitigation , a report based on responses from 1,625 IT and security leaders.
Rubrik’s research points to a growing concern among organisations deploying AI systems that can access data, interact with software and initiate business workflows.
Continue reading
