Insider incidents cost organisations $2.4 million a year, and AI agents now cause more than a quarter of them

Insider-related security incidents cost organisations an average of $2.4 million over the past 12 months, according to Trellix’s Mind of the CISO: The New Insider report. The research, carried out by Vanson Bourne, surveyed 600 chief information security officers across the Americas, Europe, the Middle East and Asia Pacific in August and September 2026. Almost every organisation in the study (98%) had at least one insider-related incident, and insider risk made up 39% of all the security incidents they dealt with.

Actions taken by AI-driven or autonomous systems caused 28% of those insider incidents on average. That’s almost level with malicious employees and trusted users at 29%. Accidental or negligent human behaviour still leads at 43%. That’s the familiar picture of someone sending a file to the wrong person, or pasting a contract into a public chatbot to save themselves twenty minutes on a Thursday afternoon.

“The growth has gone to a level that is unprecedented. We haven’t seen this level of growth before. Now the technology has to catch up, and regulation has to catch up, which is always an afterthought,” said Vibin Shaju, Vice President, EMEA Solutions Engineering, Trellix, in an interview with The Source Code.

Agents behave like staff with no badge

“It’s like installing an app on your mobile phone. Suddenly it asks for permission to use the camera, access your files, and most users just want the app to work. It’s the same with AI agents. We want the agent to work, so the permissions are not really monitored,” said Shaju.

About 78% of CISOs now count actions taken by AI agents, autonomous systems or machine identities as insider risk. That’s a bigger share than those who count malicious employees (75%) or accidental human error (70%). Think about what an agent actually does. It logs in, reaches sensitive systems and acts on someone’s behalf, yet many agents aren’t tagged in any directory. So when something goes wrong, it’s hard to say what the agent did or who asked it to.

Some 95% of organisations in the study had an incident linked to an AI or autonomous system in the past year, and 63% say they can’t see enough of how employees use AI tools. The regional figures show how quickly that can change. In the Middle East, Trellix surveyed 40 CISOs in the UAE and 40 in Saudi Arabia. Costs there averaged $2.41 million, and autonomous systems already cause one in four insider incidents.

“Everyone started to embrace AI. The business needs it, HR needs it, finance needs it. Now we’re asking what’s going to happen on the visibility side,” said Shaju.

Every slow hour adds to the bill

On average, organisations take seven hours to detect an insider-related incident and another eight hours to respond, according to Trellix. Fifteen hours sounds manageable until you think about what an agent with broad permissions can do in that time, or how fast attackers now move.

“Everything is happening in a matter of minutes to hours. By the time we see a new vulnerability in the organisation, there is already an exploit made against it, and it is already looking for organisations that have that vulnerability,” said Shaju.

The Ponemon Institute’s 2026 Cost of Insider Risks Global Report studied 354 organisations that had material insider incidents. It counts costs differently and puts the annual figure much higher, at $19.5 million, but it also shows how directly time turns into money. Incidents contained in under 30 days cost $14.2 million a year on average, while those that dragged past 90 days reached $21.9 million. Containment was the most expensive stage, taking an average of 67 days and $247,587 per incident.

In the Trellix study, a third of CISOs (33%) name too many alerts or false positives as a barrier, and 79% agree that AI and autonomous tools are making investigations more complex and slower. In practice, analysts handle the urgent alerts and the rest get closed unread or skipped.

“I know some of what is happening, but a lot of risky behaviour is getting through. We are chasing the wrong people and chasing false positives,” said a chief security officer at a public sector organisation in Colombia, responding to the survey.

“What we’ve seen now is that attackers have started to exploit medium-level attack methods. They don’t generate a lot of high-severity alerts in the SIEM, but a combination of medium-severity alerts can eventually end up in a big incident,” said Shaju. “Tracing a malicious IP address is normally done by a tier-three analyst. He connects to the IPS, finds that information, connects to the firewall, and goes to that machine. It takes hours and hours. So the question is how I make sure every single log in my organisation gets triaged without a huge human cost,” he added.

Boards are watching the AI meter

Uber used up its entire 2026 budget for AI coding tools in about four months, according to Fortune. “That link is not there yet,” said Andrew Macdonald, Uber’s president and chief operating officer, on the Rapid Response podcast. He was talking about the connection between rising use of the tools and new features for consumers.

“The board is using AI on a daily basis, and everyone is seeing in the news what has happened to bigger organisations, from a threat point of view and in how some of them failed to predict the cost of AI. So the board now has a concern. What is my operating cost when I bring in all of these AI tools, how do I consolidate them, and how do I secure them?” said Shaju.

Budget limits or implementation costs hold back 30% of CISOs from using AI or automation against insider risk, Trellix found. The bigger barriers are lack of trust in AI decision-making (40%) and limited in-house skills (39%). For security teams, costs add up twice: once on the agents the business adopts, and again on the AI they need to monitor them.

“For the agentic SOC, sending sensitive security telemetry to a public cloud, absorbing unpredictable token costs, operating without visibility into how AI decisions are made: those are real barriers. An agentic SOC has to work on the organisation’s terms,” said David Soto, Chief Information Security Officer, Trellix, in the report.

“I cannot go into a defence organisation and say you need a cloud-based AI model. They don’t have anything on the cloud, so how are they going to use that model? Everyone needs to understand what model they can adopt,” said Shaju. Trellix recently released Trellix Wise for Private Cloud for organisations that need to run AI-driven security operations inside their own environment.

CISOs now have to explain the money

Some 82% of CISOs say senior leaders want clearer reporting and metrics on insider risk than they can provide today. In addition, 74% say their board doesn’t clearly understand the risk or what it could cost, and 40% struggle to put a number on the business impact. Splunk’s 2026 CISO Report found the same strain: 85% of CISOs said low cybersecurity knowledge among non-technical executives makes working with them harder, and 41% said they can’t directly link return on investment to their risk work.

“The conversation used to be about whether AI belonged in the SOC, and now it is about control: who can see what AI is doing, who can act on it, and how fast,” said Siju Ramachandradasan, Vice President, Customer Success and Professional Services EMEA, Trellix.

Almost all CISOs (99%) say AI and automation will be critically or very important to managing insider risk over the next 12 to 24 months, and 41% plan to prioritise investment in detecting and responding to risks from AI and non-human identities. So the case they take to the board is about productivity: the same team handles every alert, and a person still signs off on the big decisions.

“With the existing resources, how can I make my tier-one SOC analysts much faster, and how can I elevate them to tier two? With a transparent AI model, I can say that based on this triaging, I have 80% confidence that this is bad; do you want to take this action? The human will still be in the loop. I don’t want to shut down an organisation based on automated intelligence,” said Shaju.

Liability is outrunning the rulebook

Some 92% of CISOs say their responsibility for managing insider risk has grown over the past year, and 69% worry about their own liability if an insider-related incident happens. Splunk’s research found 78% of CISOs concerned about personal liability, up from 56% the year before.

“Somehow, when anything goes bad in technology, it becomes a CISO problem. In the case of AI, the adoption happens in the business unit, so it is an HR problem; it’s a finance problem. We cannot blame an agent for an incident. We can only hold the people responsible for the adoption and implementation of that AI model. You can’t bring a model to court,” said Shaju.

“A nightmare” is how a US-based CISO at a financial operations provider described the patchwork of state requirements in an interview for the Trellix report. They called instead for a “unified matrix of regulations” that would help organisations “safely use AI for legitimate business purposes.” Almost all CISOs in the survey (98%) see a role for governments or regulators. Most of them want practical guidance and clearer requirements, and fewer want tougher penalties.

“If you don’t give them an application, they will find their own way to get one, which is also bad for you. So you have to adopt it, but you have to adopt it in a secure way,” said Shaju.

Sindhu V Kashyap

Global Technology Journalist & Multimedia Storyteller | Covering Founders, Investors & Leaders Reshaping Tech | Writer · Interviewer · Moderator · Editor

Next
Next

Half of workers have connected AI tools to company systems without IT. Security now has to tell the employee from the agent