Episode 2 of the Infiligence Pulse podcast, in collaboration with Control Alt. Host Vida is joined by Infiligence CEO Ram Sathia and systems validation expert Bill Gibson.
What is discussed?
- Why traditional validation breaks down with non-deterministic AI?
- Proportional governance: the FDA's two-axis risk model (AI authority vs. consequence) and CDRH's draft guidance on GenAI medical devices.
- Human-adjacent frameworks, including onboarding a large language model (LLM) as if it were a new employee, implementing competency-based graduation, and distinguishing between production and pre-production models.
- Reusing ALCOA+ and 21 CFR Part 11 as AI compliance dashboards instead of reinventing validation.
- The client data training paradox — why training on client data mirrors training human staff?
- Over-governing, shadow AI, and the cost of governance roadblocks.
The CXO Playbook
The CXO playbook emphasizes risk-based "just enough" validation and the use of token budgets.
Key takeaway:
Governance belongs inside the workflow rather than at the end gate.
