MIT AI Explains Why Their Agents Lie and Cheat.
A recent MIT Tech Review article explored how AI agents lie. It’s known as “reward hacking”, or when AI agents take the fastest path to winning by any means necessary. The stronger the model, the sneakier it can be.
In 2016, an AI played a racing video game where it kept driving around to collect bonus points. In July 2026, OpenAI used several undisclosed vulnerabilities on Hugging Face’s testing server to reach the correct solution.
With increasingly capable AI comes increased ability to obfuscate its behavior. This is a huge problem when building autonomous agents that will operate in the physical world. The AI isn’t being malicious, but it’s totally subverting the goal.
Source: www.technologyreview.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides
Author
Evgenii Arsentev
PhD · Chief Executive Officer, digital health
Articles · Latest articles