Ignacio Lopez has one rule for every client: AI never does the math. AI reads, code decides, a person signs, and the logs prove it.
In this episode of the AntiCheckbox Podcast from the GRC Engineering Club, Omar Sangurima talks with Ignacio Lopez, a former KPMG auditor (SOX and SOC reporting) and JP Morgan information risk manager who now helps wealth management, legal and real estate firms roll out AI in highly regulated environments. He is based in Miami and has been through central bank audits in Argentina that decided how many branches a bank could open.
Ignacio explains the difference between probabilistic AI and deterministic code, and why anything involving money has to be deterministic, traceable and explainable. He walks through the controls he actually implements: an AI policy before any tool, why a prompt is a request and not a control, connector policies and read-only permissions, incident management, dual controls, and naming one human who is responsible for every AI output. He also shares a real invoice automation for a property manager, where AI reads and code calculates, and how he backtests it against years of history.
KEY TAKEAWAYS
The four-part sentence Ignacio wants every listener to remember.
Why "it looks right" is the most dangerous thing about AI spreadsheets.
What a prompt is, if it is not a control.
The permission most people forget an AI connector inherits.
The one control that changed how his clients review AI output.
How he uses years of past data to test an AI workflow before it goes live.
Which industries are adopting AI slowest, and why that is an opportunity.
FIND IGNACIO
LinkedIn: https://www.linkedin.com/in/ignaciolopez2017/
Send him one workflow: https://work-smart.ai/send-me-one-workflow?utm_source=anticheckbox&utm_medium=podcast&utm_campaign=appearance
Get his three-control checklist: email ignacio@work-smart.ai with "checklist" in the subject
CHAPTERS
0:00 Who is Ignacio Lopez?
2:09 AI never does the math
2:45 From KPMG auditor to AI implementation
6:01 Probabilistic AI vs. deterministic code
8:43 The cost of using AI where a script will do
10:02 Explainability: why black boxes fail audits
12:07 A prompt is not a control
14:57 AI adoption is a culture change
16:58 AI compliance for SEC registered advisors
17:26 Is your data ready for AI?
21:11 AI connectors and least privilege
24:46 Lessons from central bank audits
29:05 Human in the loop and a named owner for AI output
33:02 Real example: AI invoice processing with code doing the math
34:36 Backtesting AI against past data
36:18 Which industries are slowest to adopt AI
40:29 Start with the policy