Evolution, Not Revolution — But Still Dangerous
Bjorn R. Watne, global chief information security officer at Interpol, described AI’s role in cybercrime as “an evolution and not a revolution.” Scam operators are using AI to run parallel attacks against multiple victims simultaneously. Improved translation tools and synthetic digital identities are eroding the visual and linguistic cues that once helped people identify fraud.
The practical result: the volume of attacks increases while the cost per attack drops. For enterprise security teams, that means the threat surface expands even when the underlying criminal tactics remain familiar. Related concerns around AI-powered hacking tools point to the same acceleration effect.
What Companies Should Actually Protect
Watne’s guidance cuts through the noise with a straightforward prioritization framework.
Start with your crown jewels. Before deploying defenses broadly, identify the assets most critical to keeping operations running. Then ask who would realistically want to steal or disrupt them.
Match defenses to actual threat actors. A company facing opportunistic criminal groups requires a different posture than one targeted by advanced persistent threat actors. Watne was direct: “There is no need for everyone to try and protect against everything all at once when they’re not being targeted by it.”
Close the boardroom gap. Many organizations still treat cybersecurity as an IT department concern rather than a strategic business risk. Watne noted that cyber risk is rising on corporate risk registers, but has not yet fully reached the boardroom at many companies. That disconnect between senior management and security strategy remains a structural vulnerability.
Agentic AI: The Risk Category Worth Watching Now
Beyond current attack patterns, Watne flagged agentic AI as a distinct and emerging concern. As AI systems gain the ability to take actions on behalf of users — booking, purchasing, operating physical systems — the consequences of errors or manipulation shift from informational to operational.
“One thing is that an AI is feeding you the wrong information,” Watne said. “But when an AI is actually performing an action and it starts doing the wrong actions, especially in the physical domain, that can have consequences on bodily harm of humans.”
He pointed specifically to AI deployed in vehicles and self-driving systems as a concrete example of where this risk becomes physical.
The Trust Problem
Watne identified an underappreciated vulnerability: people extend far more trust to technology than to financial services. In banking, users expect friction — PIN codes, verification steps, fraud alerts. With new apps and devices, the default behavior is to click through permissions quickly and grant broad access without equivalent scrutiny.
As agentic AI systems acquire greater access to accounts, calendars, communications, and physical devices, that habitual trust becomes a meaningful attack surface.
The Practical Takeaway for Security Teams
Three priorities emerge from Watne’s analysis:
- Asset mapping first. Know what you are protecting before deciding how to protect it.
- Threat-model your defenses. Align security investments to the actual adversaries relevant to your industry and data profile.
- Audit agentic AI deployments now. Any AI system with the ability to act — not just advise — warrants a higher level of access control and oversight than most organizations currently apply.
The speed and scale advantages AI gives attackers are real. The response is not to defend everything simultaneously, but to defend the right things with precision for stronger fraud prevention.
Comments (0) No comments yet
Want to join this discussion? Login or Register.
No comments yet. Be the first to share your thoughts!