Loophole: Adversarial Agents To Stress Test Your Morality — Brendan Rappazzo, Morgan Stanley
Source
youtube.com
Author
AI Engineer
Date
Context
Loophole uses opposing AI agents to expose gaps between a person's values and the rules written to express them. One AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → turns plain-language moral preferences into a formal code. Two others search for different failures: behavior the person opposes but the code permits, and behavior the person accepts but the code forbids. A judging agent attempts translation repairs and sends unresolved conflicts back to the person. The talk presents this as a way to generate cases someone might not have anticipated.
Its useful output is a concrete disagreement to examine. Finding such a case can reveal whether a rule needs clearer wording or whether the person needs to resolve an ambiguity in their own preferences.
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
videoHow to Write Your First AI Eval - SallyAnn DeLucia, Arize AI
videoBuild Your Own Team of Managed Agents - Jack Wotherspoon, Google DeepMind
videoHow to Build an AI Code Review Agent - Daksh Gupta, Greptile
videoHow to Build Voice Agents That Don't Fall Apart When You Interrupt - Jesse Hall, LiveKit
