AI Agents Aren't Rogue. They're Relentless.
Techstrong TV Interviews
•
17m
Doing Exactly What They're Told
AI agents are often described as going rogue. In reality, most are doing exactly what they were asked to do. They just pursue the goal more creatively than people would, and they never get tired.
Bitdefender Director of Cybersecurity Services Nick Jackson examines that misconception with Mike Vizard. Jackson compares it to asking someone to wash your car. A regular cleaner knows which sponge and soap you prefer. A stranger will use whatever works. Without detailed instructions, agents find their own paths, including backdoors no one knew existed.
Non-Human Identities Need More Control
Jackson sees AI agents as non-human identities inside the business. He argues they need tighter controls than human workers. Model providers keep adding guardrails as they learn what to restrict. Organizations still have to manage how agents run in their own environments.
Regulation will take time, much as it did with cloud. Organizations can act now, though. Bitdefender research found about 47% of organizations lack visibility into the AI tools employees use.
Unlike most technology, AI often enters through legal, HR or marketing rather than IT. Some teams use AI features they don't even know were added.
Fundamentals Still Matter
Most of the needed controls are security basics. Change management, procurement reviews and privacy checks would catch many issues. Jackson cites one customer about to send EU data into a US-hosted model without considering GDPR.
Enterprises are moving faster than in the past. Jackson worries more about smaller firms and consumers. He also asks how new professionals will learn the basics if AI handles every mundane task.
Inform the Risk, Don't Block It
Security's role is to identify risk and inform the business, Jackson says. The business then owns the decision.
Risks include agents acting unexpectedly, stolen credentials and tampered agents in multi-agent chains. Jackson recommends time-bound access, regular skill checks and human oversight. Non-human identities now outnumber people online, so teams need to get ahead of it.
Up Next in Techstrong TV Interviews
-
The Sticker Shock of AI Infrastructur...
Governance Meets the AI Era
AI infrastructure costs are about to climb, and many organizations don't see it coming. AI agents will consume far more resources than today's workloads. Most teams only notice when the first bill arrives.CloudBolt CEO Alexander Thomson examines that challenge with M...
-
Anaconda Puts Guardrails on Agent Swarms
AI agents now work side by side in swarms, and that raises new security questions. Alan Shimel talks with David DeSanto, CEO of Anaconda, about how the company brings guardrails, least privilege access and visibility to agent swarms.
From Python packages to an AI platform
DeSanto joined A...
-
Aviatrix Makes Breach Containment Fre...
Security teams cannot patch fast enough to keep up with this month's flood of vulnerabilities. Alan Shimel talks with Doug Merritt, CEO of Aviatrix, about Clearance and why breach containment is now the most reliable way to limit damage. Aviatrix is also making Clearance free for any enterprise t...