SpeakersAgendaWhy attendPartnersVenue
Locationsā–¼

Get invited

Register

Partnerships

Request your invite

Call to action
Your text goes here. Insert your content, thoughts, or information in this space.
Button

Back to speakers

Aaron
Portnoy
Chief Product Officer
Mindgard
Aaron Portnoy is Chief Product Officer at Mindgard and the inaugural Hacker Fellow at Dartmouth College, applying over twenty years of offensive research to AI security. He created the Pwn2Own hacking competition, organizing and judging its first six iterations while running research at TippingPoint's Zero Day Initiative, and went on to co-found Exodus Intelligence, one of the first firms to commercialize zero-day research. Over his career he has personally discovered hundreds of zero-day vulnerabilities in software from vendors including OpenAI, Cursor, NVIDIA, Microsoft, Google, Amazon, Palo Alto Networks, and Adobe; authored the award-winning IDA Toolbag; and published research in Phrack. He has led offensive programs at Raytheon and Boldend and pioneered attack surface management research at Randori through its IBM acquisition. His current research focuses on the security of AI systems. Featured in TIME Magazine's 2014 cover story "World War Zero," Aaron has delivered over thirty invited talks at venues including Black Hat, REcon, CanSecWest, EkoParty, USENIX WOOT, BlueHat, RSA, and the NSA Distinguished Speaker Series.
Button
04 June 2026 12:00 - 12:30
The Emperor's New Guardrails
Every vendor selling enterprise AI talks about their guardrails. Every CISO is being told the controls are in place. How convinced are you that they actually work? The benchmarks are academic. The vendor numbers lack methodology. The real-world threat data is still too thin to calibrate against. Anecdotally, we've yet to encounter a guardrail that we couldn't break. Drawing on over 20 years of offensive security experience and having discovered 80+ vulnerabilities across leading AI products, this talk shows what is actually keeping (and not keeping) your AI systems safe. The practical path forward is to test continuously, treat every vendor claim as provisional, and build the evidence base in-house. We'll cover what we've learned while compromising defensive AI technology, and how security leaders can gain meaningful assurance today.