Third Party Threat Hunters podcast

Third-Party AI Risk, Explained with Shankar Krishnan

0:00
26:43
Spola tillbaka 15 sekunder
Spola framåt 15 sekunder

Your third-party vendors are not “adding AI” as a feature. They’re turning into fast-changing agent stacks with models, connectors, plugins, skills, MCP servers, and subprocessors that can shift underneath you in days or even minutes. That’s the uncomfortable backdrop for our conversation with Shankar Krishnan, co-founder of Prompt Armor and a longtime AI threat researcher, as we unpack what third-party AI risk really looks like when the attack surface won’t sit still.

We get concrete about the threat that keeps creeping into everyday workflows: indirect prompt injection. Malicious instructions can hide in normal places like documents, emails, and web pages, then get “executed” by an LLM agent that has access to confidential data and powerful actions. We talk through why the scariest outcome is prompt injection leading to data exfiltration, how shadow AI shows up inside the supply chain when vendors enable AI by default, and why the shared responsibility model increasingly pushes risk decisions onto customer configuration.

From there, we zoom out to the program level: why SOC 2 reports and standard security questionnaires still have value but fail to describe AI agent behavior, rapid model changes, or evolving dependencies. We cover practical steps for continuous monitoring, tighter alignment between third-party risk teams and AI governance, and the contract clauses that matter most when vendors can change models, features, and training practices without slowing down. We also dig into nth-party mapping and concentration risk, where a handful of foundation model providers can quietly become critical dependencies across your vendor ecosystem.

If you’re responsible for AI vendor security, supply chain security, or third-party risk management, this one will help you reset your mental model and prioritize what to monitor next. Subscribe, share this with your risk or security team, and leave a review, then tell us: what’s the hardest AI vendor change for you to track?

Send us a question for a future episode

Support the show

Fler avsnitt från "Third Party Threat Hunters"