How safe is your organization from cyberattacks? You might invest in the latest defensive capabilities, carry out regular audits to identify weak spots, and keep your staff trained and up to speed with the most recent developments. But if someone inside the building gives away the keys to the kingdom, it’s all for naught.
Despite the rapid pace of malware development and the onset of AI tools that can augment that process even further, this threat from inside can play an inordinate role in cracking even the toughest of nuts. After all, safeguards are meaningless if they can be bypassed by someone who is ostensibly batting for the home team, whether accidentally or deliberately.
To find out the kind of damage that can be done in these situations, and how you can protect yourself from the insider threat, we spoke to a range of cybersecurity experts who shared their thoughts on the best ways to keep the danger at bay. In some cases, it could require a total rethink of your defensive posture.
The ‘direct path’ inside an organization
“Internal vulnerabilities are where attackers turn a normal account compromise into a privileged account compromise,” according to Erich Kron, CISO Advisor for security experts KnowBe4. His statement neatly highlights the dangers posed by insider threats within organizations. This is because, unlike traditional breaches, insider compromises originate with internal company accounts and credentials that have already been afforded a degree of trust.
Michael Leland, VP and Field CTO at enterprise AI firm Island, agrees that the insider threat is a serious one. Put simply, Leland said, “Insiders can be malicious, negligent, or simply compromised, and either way, they remain one of the most direct paths attackers use to get inside an organization.”
The very nature of their roles as trusted employees at the company makes them a potential danger from the start. As John Podboy, CISO at cyber defense firm Semperis, told us, “The biggest weakness is trust itself. Organizations give people the access they need to do their jobs, but excessive privileges, stale accounts and third-party access can create opportunities for attackers.”
Sometimes, those opportunities arise when an employee is secretly working with adversarial groups. But hackers don’t always need a double-crossing contact feeding them logins in order to wreak havoc. Sometimes, the “insider” is actually just the credentials themselves, which can be leaked through carelessness, exfiltrated via a phishing attack, or purchased on the dark web. Leland went on to say that “[a] single set of stolen or misused credentials can give an outside attacker the same reach as a trusted employee.”
Dave Lewis, Global Advisory CISO at identity security company 1Password, made a similar point. “The classic, and rather tired, image of the insider threat is the disgruntled employee walking out with a customer list,” he said. “Now that still happens, but the bigger risk today is the employee who becomes an attacker’s access point without their knowledge.” Lewis added that “nearly every major breach of the last several years involved a valid identity at some point.”
Whatever the cause, Leland warned that any form of insider risk is particularly dangerous for organizations. This is because “insiders’ actions often look completely normal to traditional security tools because there’s no malware signature or unauthorized login to flag. It’s simply someone, or something, doing what it was technically allowed to do.”
How AI is elevating the risks
It’s that combination of innocent-looking actions and privileged access that can make insider threats so difficult to spot and respond to. And with AI toolsets being increasingly employed by organizations that might not have fully assessed the risks of doing so, the incident rate could be about to intensify.
Not only do many AI agents have access “exceeding what any human employee would get,” Leland warned, but their actions have been moved to a place where network-based detection tools often lack visibility. Think about how many AI agents can access “the user’s file system, locally-installed LLMs, and even decentralized MCP servers,” Leland posited, “all of which sit outside the reach of traditional SASE or cloud-only security stacks.” AI “hasn’t just added a new insider,” he continued. “It has exposed a structural gap in where and how insider risk needs to be monitored going forward.”
Throw in the incredible speed and scale at which AI agents can operate (and leak data) and you have a potential recipe for disaster if appropriate guardrails are not put in place.
And that’s not something that’s going to go away any time soon. In fact, “the threat will be expanding as agentic personal assistants continue to be rolled out within organizations,” said Kron. “Each agent that has access to the same things the employee does is something that must be closely monitored as these agents can be socially engineered, or can act in unexpected ways just due to their nature.”
AI is also making it easier for attackers to pose as insiders. “Voice cloning means the call to your help desk sounds like your CFO,” Lewis said. The result, he argued, is that “trust based on how someone sounds, looks, or writes is no longer a valid control. Verification has to be cryptographic and procedural.”
Until organizations “shift their mindset from reactive to proactive to ensure the business can continue to operate,” insider threats will continue to endanger companies and their defenders, Podboy cautioned. AI is radically transforming the entire industry, including the danger posed by internal leaks and breaches, and taking a passive approach simply won’t cut it.
How to keep insider threats out
The biggest problem when it comes to this issue is “overprivileged access,” Leland explained. And this is a vulnerability that isn’t limited to humans: “Employees and now AI agents routinely hold permissions far beyond what their roles require, which can turn any single compromised account into a much bigger problem than it should be.”
And because companies have limited visibility into what employees and AI agents do, from taking screenshots to uploading files to cloud storage, “you get many ways for legitimate access to quietly turn into data exfiltration,” Leland argued. “Many organizations also lack a clear mapping of who and what can touch sensitive data in the first place, so by the time unusual activity sets off an alarm, it’s often already too late.”
That exfiltration doesn’t even have to be deliberate; plenty of breaches happen from messages being sent to the wrong person, workers using unapproved apps, or sensitive files being shared beyond their scope. In situations like that, “a strong Data Leakage Prevention system is critical to the ingress and egress of the organization,” Kron advised, before adding that “tools that identify possibly misaddressed email messages should be in place.”
Regardless of whether the act is deliberate or not, an effective defense should focus on the action, not the intent, Leland told CISO Voice: “The strongest defenses don’t try to distinguish intent in real time. They apply consistent least-privilege and monitoring controls that catch the action itself, whether it was a mistake or a plan.” Updating training, education, and company policies should also be a priority, Kron urged.
While this is far from an unsolvable problem, fixes require careful and determined action. Companies must “enforce least-privilege access tied to real identity and role, monitor and log activity continuously rather than trusting access once granted, and require approval for high-impact actions involving sensitive data,” Leland advised. Disgruntled employees, those who travel often, and those who are about to leave the firm should also be “monitored for unusual behavior,” Kron added.
These controls must live at the ‘last mile’ where users come into contact with data, Leland noted, and organizations should engage in red teaming and access reviews to highlight potential issues before they can be exploited.
Lewis said orphaned accounts are “the easiest footholds because nobody is watching them.” In one access assessment, he “found 10 super user accounts that belonged to former employees. One account was active in the prior calendar year, but the individual had been deceased for 5 years.”
Striking the balance between safety and efficiency can be tricky, and it’s likely that many people reading this will be wary of disrupting their teams unduly. But as Leland put it, “the goal isn’t to eliminate access or slow teams down. It’s to make sure every permission granted is deliberate, scoped, and observable.” The fewer gaps you have, the fewer chances there are for privileged account compromises to rear their ugly heads.