The Morality Machine — Episode 3: The Ethics of LLM Poisoning

Welcome to the Morality Machine, a podcast created by Elise Hachfeld and Daniel Evans for our Computing Ethics class.

In this episode, we explore the dark side of large language models (LLMs.) Any time a new technology is introduced, a race begins to hack it. It’s no different for AI. Both bad actors and activists have discovered how to ‘poison’ LLMs to impact their performance.

---

Overview

First, we examine one potential dark side of AI poisoning: Russia’s use of misinformation and disinformation campaigns and the Pravda network. Next, we discuss the software Glaze and Nightshade, which is a positive example of how AI poisoning can be used to protect intellectual property.

Listen to the full episode below:

Sources

1. Russia seeds chatbots with lies. Any bad actor could game AI the same way.
2. Russia-linked Pravda network cited on Wikipedia, LLMs, and X
3. This tool strips away anti-AI protections from digital art
4. The AI lab waging a guerrilla war over exploitative AI
5. A small number of samples can poison LLMs of any size