Approving a foundation model is not the same as governing what teams build with it. The approval covers one decision. Then come the agents, the integrations, and the dependencies teams add every week, and few of those ever reach a security review. In the premiere of Artificial Trust: Exposing Blindspots in Today’s Enterprise, ActiveState CEO Abby Kearns and Nicholas Giedt, Founder and Fractional CISO at Trellis Security Advisors, talk through how security leaders set trust boundaries, control what enters the stack, and protect developer speed without inheriting a remediation bill that grows faster than anyone can pay it down.
What You’ll Learn
- Why approving a foundation model tells you nothing about the agents and tools teams build on it
- What a SOC 2 report and a penetration test can and can’t tell you about an AI vendor
- How a trusted AI vendor becomes your newest risk when secret scanning stops at their perimeter
- How Simon Willison’s lethal trifecta (private data, untrusted content, outbound communication) shows you which capabilities to keep apart in an agent
- How capability-based, short-lived access keeps AI agents from holding standing permissions
- How to review AI-generated code at speed when a single pull request runs to 20,000 lines
Nicholas Giedt is the Founder, Principal, and Fractional CISO at Trellis Security Advisors, where he helps companies roll out and secure AI so they can ship it instead of shelving it. Before advising, Nicholas built and scaled security programs at Strava and Roblox before joining Trellis Security Advisors.
If you enjoyed this episode, make sure to subscribe, rate, and review on Apple Podcasts, Spotify, and YouTube Podcasts. Instructions on how to do this are here.
Key Highlights - [00:00] Intro
- [02:02] Why Companies Shelve the AI They Paid For
- [04:13] Compliance Is the Floor
- [06:16] How AI Changed Third-Party Risk Reviews
- [12:53] Trust Boundaries and Citizen Developers
- [19:22] Data Classification and Simon Willison’s Lethal Trifecta
- [25:59] Who Reviews a 20,000-Line Pull Request?
- [30:58] Incentives, Guardrails, and Shifting Review to the Prompt
- [34:37] From Identity-Based to Capability-Based Access
- [40:12] What Changes in the Next 6 to 12 Months
- [43:48] Who’s Accountable When Agents Go Off the Rails
Episode Resources:- Nicholas Giedt on LinkedIn
- Trellis Security Advisors Website
- Abby Kearns on LinkedIn
- ActiveState Website
Artificial Trust: Exposing Blindspots in Today’s Enterprise is handcrafted by our friends over at: fame.so