Defense in Depth How Best to Prepare Your Data for Your Tools
16 snips
Jan 22, 2026 Matt Goodrich, Director of Information Security at Alteryx, shares insights on AI data hygiene and governance. He highlights the challenges of relying on polished AI outputs, warning about their potential misleading nature. Goodrich emphasizes the importance of integrating human oversight in AI workflows and proposes using multiple AIs for cross-verification. He advocates for treating AI outputs as research rather than authoritative, pushing for a skeptical approach, and underscoring the need for governance to establish trust in AI-driven security systems.
AI Snips
Chapters
Transcript
Episode notes
Keep Humans In The Loop Initially
- Put humans into AI workflows where stakes are high and remove them only after trust is proven.
- Build citation, double-verification, and escalation paths before reducing human oversight.
Use Multiple Agents As Skeptical Checkers
- Using multiple AI agents enables one agent to skeptically check another, reducing hallucination risk.
- Split problems across agents so independent checks can catch misleading directions.
Non-Determinism Means Ongoing Scrutiny
- AI is non-deterministic so hallucinations are inevitable even with flawless inputs.
- Constantly challenge outputs and reasoning, and verify underlying sources of truth.