News of rogue AI agents breaking out of controlled environments is becoming frequent. In this episode, we asked Prateek Bhajanka for his views. He is a seasoned cybersecurity strategist, a former Gartner Analyst and Magic Quadrant author with a track record of 15 years in research, engineering, and consulting.
In the past few weeks, there have been news reports about agents created by Anthropic, OpenAI and Google AI that broke out of their testing environments to breach websites and gather information. These agents performed beyond their authority and were labeled ‘rogue agents’.
While their actions may seem harmless today, people are already talking about the existential threat to humanity. And what might happen when you put advanced versions of these models into physical robots.
“Looking at the incidents that have happened … I wouldn’t be surprised if the threat of AI agents that we are anticipating or that we are predicting is real. There are many industry conversations about the regulated use of the AI and the regulated development of frontier AI models – because of the way things are pacing,” said Prateek Bhajanka, CEO and Chief Field CISO at FieldCISO Advisory.
Prateek is convinced that frontier AI companies are deploying guardrails and controls to ensure that the frontier AI capabilities of their models are not misused for “unconstitutional purposes, risky and security [behavior], hacking purposes or any kind of illegal activity.”
“But as you can imagine, no technology provider can factor in all combinations to make sure that all the possible scenarios have been taken into account and are controlled,” said Bhajanka.
In this video podcast, Prateek discusses:
- Observations on rogue AI agents and their actions;
- Controls for limiting actions of agents;
- Measures that security vendors should be taking.
Prateek is a seasoned cybersecurity strategist. A former Gartner Analyst and Magic Quadrant author, he brings over 15 years of diverse experience across research, engineering, and consulting for industry leaders like Accenture, Deloitte, SentinelOne. A regular speaker at global forums, Prateek is dedicated to helping organizations navigate the complexities of Cybersecurity Maturity and Ransomware Defense, AI Governance, while advising security vendors on high-impact Product and GTM strategy.
