Simon Willison on AI Agents and Harder Work
Simon Willison explains why AI agents leave him only hard decisions, how he is rethinking prompt injection, and why he urges developers to build weirder things.
Simon Willison explains why AI agents leave him only hard decisions, how he is rethinking prompt injection, and why he urges developers to build weirder things.
Chrome’s engineers built a synthetic web where browser agents can be tested without real-world risk, including thousands of simulated sites and prompt-injection attacks.
Google’s Chrome team built a sealed, fake internet called the Mirror Dimension to test browsing agents built on Gemini against attacks such as prompt injection.
Prompt injection reached three major coding agents in 2026. The lesson for AI agent security: limit what a fooled run can do, because you can’t stop the fooling.