A study by API management platform Graviteesuggests that more than half of the three million AI agentscurrently in use by organizations in the US and UK are “not properly vetted.” Based on a December 2025 survey conducted by Opinion Matters of 750 executives and IT professionals, the results revealed that AI agents are evolving faster than security teams can keep up.
According to Rory Blundell, CEO of Gravitee, there are now over three million AI agents operating within companies, which he described as a workforce larger than Walmart’s total global workforce. The three million figure is based on an extrapolation of the survey results, based on government estimates for 8,250 businesses in the UK and 77,000 businesses in the US that employ 250 or more employees.
The average number of AI agents deployed per organization is 36.9. When respondents were asked if their organization “experienced or suspected a data privacy related to AI agents in the past 12 months,” 88% responded that they had.
See also: Docker fixes critical flaw in Ask Gordon AI
The average percentage of agents who are not monitored and uninsured, according to the findings, was 53%.
When asked how the study came about, Blundell wrote in an email, “We’re all familiar with stories of AI agents going unchecked: deleting codebases, leaking confidential information, inventing false data. The working hypothesis that prompted this research was that, while AI agent development is reaching an exciting stage, businesses have yet to get to grips with managing AI agents. The research confirms this.”
AI agents: Insufficient control is a global problem
Agents, he said, “can offer businesses a huge productivity boost, but we need to be realistic about the risks: without governance and oversight, they can easily start to become liabilities and a risk for consumers and businesses.” Furthermore, Blundell said that, despite the fact that respondents were only from the UK and US, “this is absolutely a global problem. Companies all over the world are using AI agents, and across the spectrum there is a gap between the level of development and the level of governance. We have a strong customer base in the EU, where we see the same problems.”
David Shipley, head of Canada’s Beauceron Security, said: “The one thing that shocks me is that people think only 53% of agents are not monitored. It’s a higher percentage.” He likened the results of the Gravitee study to a “lesson about the Titanic, which everyone in technology continues to ignore. The Titanic disaster didn’t happen because they didn’t know there would be icebergs on the voyage. They knew it was peak iceberg season, they knew they were going very fast.” Shipley said the ship’s captain and crew “thought they would detect [an iceberg]. If they didn’t and they hit one, they thought the technological controls would protect them to help them recover.”
See also: AI strengthens the attack chain in AWS environments
They trusted the so-called waterproof compartments that, as it turned out, were not waterproof, but, more importantly, they trusted the new wireless communications technology that they could use to call for help if they were in trouble.
The equivalent today: "Informatics and security can save us if we have a problem with our AI agents. Wrong then, terribly wrong now," he said.
Shipley continued by saying that “we know that AI agents are inherently dangerous and untrustworthy. There is literally mathematical proof that shows this. So, we know that there are icebergs. Let me repeat that for those in the back of the room: 100% of AI agents have the potential to become uncontrollable. If a vendor assures you that it is not possible and their underlying technology is an LLM, they are lying. We know that we are moving too quickly in adoption for the risks that we know exist.”
Shipley added, “Now, the funny part: imagine if the Titanic made the same choices, knowing that the watertight compartments were not working (i.e. missing monitoring for 53% of AI agents). We know that by the time IT and security intervene in an AI agent-related risk, the damage is already done (the ship sinks very quickly).
The real issue is invisible AI agents
Manish Jain, research director at Info-Tech Research Group, said that as the "exponential" pace of AI development continues, there are expected to be more AI agents worldwide than human workers by the year 2028.
🔒 Protect your privacy with Proton VPN
Swiss VPN from the creators of Proton Mail — strict no-logs policy, strong encryption, and built-in NetShield that blocks ads, trackers, & malware.
- ✔ No-logs, based in Switzerland (except 14-Eyes)
- ✔ NetShield: blocks ads, trackers & malicious domains
- ✔ Covers all devices — free version available
The link is an affiliate link — SecNews may receive a commission at no additional cost to you. It does not affect the independence of our article writing.
“It will be one of the biggest challenges for business and IT executives to manage them without limiting the innovation that these AI agents bring with them,” he said. Even today, he noted, “we see that most AI agents in enterprises operate without supervision. Many organizations don’t even know how many agents they have, where they operate, or what they can touch. If you don’t know how many mules are in the barn, don’t be surprised when one kicks in the door.”
See also: Japan and Britain strengthen cooperation on cybersecurity
Jain pointed out that AI agents are no different. “Unregistered agents often emerge through low-code tools and informal experiments, bypassing traditional IT control until something breaks. You can’t manage what you can’t see. So we need to understand that the real issue isn’t ‘uncontrolled AI agents,’ it’s invisible AI agents.”
AI agents are no longer useful bots. They often operate with authorized but broad credentials, persistent access, and undefined accountability. This can be a costly mistake as overly privileged agents are the new insider threat. We need to define access levels for AI agents.
