Summary
The panel rejected “zero tolerance” as a deployable AI standard, arguing that high-stakes domains should minimize, govern, and intelligently absorb risk rather than pretend it disappears. Speakers emphasized guardrails, backtesting, early deviation signals, board-level questioning, and transparent sharing of failed experiments as the path to trusted AI in insurance, fraud, healthcare, and defense.
Key Topics
5 key topics from Spencer Schaefer, Karen Silverman, Gabriel Banks and Barbara Junge at HumanX Amsterdam 2026. Thicker branches were mentioned more often in the talk.
Risk, Responsibility & Trusted AI in Zero-Tolerance Environments
Click a topic to open it
Key Takeaways
- Design governance layers before deploying high-stakes AI.
- Ask what projects were killed before production.
- Share failures transparently to build stakeholder trust.
- Use backtesting and guardrails to reduce avoidable risk.
- Treat zero tolerance as an aim, not an operating assumption.
“those failures and saving those failures and sharing those failures is actually what's building more trust”
Share
Up Next
SAN
WatchNext in agenda
An Agentic Backbone
Steve Lucas, Arun Chandra, Nidhi Aggarwal, Tomas Statius
agentic aiai governance
DEN
WatchAlso on ai governance
The Ethics Tax: Can You Afford to Build a "Good" Model?
Dylan Fox, Emre Kazim, Nina Müller
ethical aiai governance
DJA
WatchAlso on ai governance
The Autonomous Finance Office
Daniel Kim, John Glasgow, Anna Tutova
autonomous financeai governance