Companies are mass-hiring AI agents, but most don't know how to stop them in time
Companies are increasingly entrusting autonomous AI agents with more tasks, but they are lagging far behind in terms of monitoring them. A survey by Gravitee reveals a worrying situation. Only six percent of technology leaders surveyed could deny a problematic AI agent access to all information systems within an hour.
The survey, which surveyed 900 leading technology professionals from the US and UK, found that fewer than a tenth of organisations have a formal person responsible for the conduct of AI agents. Even more strikingly, on average, only 51% of AI agents have rules in place that actually limit their permitted actions.
The consequences are no longer just theoretical. More than half of respondents have experienced at least one security incident involving AI agents in the past year. Reported cases include the disclosure of login credentials, internal pricing lists, and customer contact information. Some agents have even shared sensitive data outside the organization.
The problem continues when it comes to determining responsibility. Only one in eight respondents could fully reconstruct what an individual agent was doing 30 days ago and with whose authority.
Gravitee CEO Rory Blundell warns that companies are increasingly allowing agents to make their own decisions without putting in place adequate oversight mechanisms. He sees the solution in clearly identifying each agent, limiting their rights and logging their activities.
Artificial intelligence is gaining more and more powers, but companies often still don't have an answer to the most basic question: who will stop it when something goes wrong.




















