AI’s Jurassic Park Period
Early in my career, I accidentally and unrecoverably changed data I was collecting for a federal investigation. Twenty years later, with the help of AI and a career’s worth of experience as a security leader, I intentionally did the same thing. Make no mistake, what my agent and I did together was dangerous. It was only because I had enough subject matter expertise in both the functional and risk issues that I could navigate it safely. We are in AI’s Jurassic Park period: no matter how clearly we define the rules, models will search for paths to completion. And they are very good at making those paths look safe, reasonable, and correct even when they violate policy or basic intuition. Designing the right control set is about allowing for the right expertise to be injected at the right time in the co-creation process so we can move quickly and safely into the next evolution.
About the Security Track
Security sessions at AI Engineer World's Fair 2026 in San Francisco.
When
Tuesday, June 30, 2026
3:20 PM - 3:40 PM·20m
Where
Track 5 · Room 2005
Capacity: 250 attendees
Sign in to add this talk to your schedule.
Speaker
Aaron Stanley
Head of Security and IT
dbt Labs
Aaron Stanley currently heads the Security and IT functions for dbt Labs, the leading platform for data analytics engineering. Originally a sysadmin for a regional ISP, he developed the forensic technology for Stroz Friedberg, built an eDiscovery engine for Apple, and ran the cybersecurity function at Twilio before joining dbt Labs in 2023. He believes that a good security team is one that incentivizes smart risk-taking, draws on a diverse set of viewpoints and experiences, prioritizes individual growth and development, and gives back to community at-large where possible. Aaron’s long term objective–his north star–is to improve the security posture of the world.