Skip to content

Posts
Problem space
Research
Resources
Book
About

The problem space · Internals · Backdoors

Align your AI Internals against backdoors

Internals: the activity inside the model
Backdoors: hidden behavior waiting for a trigger

Previous in InternalsInternals against sycophancyComing soon

Next in InternalsInternals against persona driftComing soon

Open research, idea stage. Nothing for sale.

Code is licensed under Apache-2.0. Content is licensed under CC BY 4.0.

Follow new posts and research by RSS. Code and tools are on GitHub.

Motion off keeps every fact and removes animation. Dark follows your device until you choose.