AI Safety Atlas

An interactive, open-source textbook on AI safety, published at ai-safety-atlas.com.

The chapters are co-authored by researchers at the French Center for AI Safety (CeSIA) and partners. The site is an Astro static build with a custom content pipeline that pulls chapter prose from Google Docs.

Quick start

git clone https://github.com/markov-root/atlas.git
cd atlas
pnpm install
pnpm dev

Open the URL pnpm prints (defaults to http://localhost:4321). No .env file is required for local development — chapter prose comes from the committed .cache/docs/ snapshot.

Commands

pnpm dev        # Start dev server
pnpm build      # Build to ./dist
pnpm check      # Fast loop: typecheck + tests (~10s)
pnpm verify     # Full pre-push gate: lint + typecheck + tests + build + smoke (~80s)
pnpm test       # Unit + integration tests only

See CONTRIBUTING.md for the full command list, env-var reference, and contribution workflow.

Documentation

Doc	What
CONTRIBUTING.md	Setup, workflow, env vars, commit conventions
TRANSLATING.md	How to contribute a translation into another language
docs/ARCHITECTURE.md	Build pipeline, BuildMode, repo layout, "why we chose this"
docs/PRINCIPLES.md	Engineering principles with code references
docs/ROADMAP.md	Where the codebase is going (Now / Next / Later / Not planned)
docs/design-system.md	Visual patterns, component catalog, spacing tokens

Contributing

External contributions welcome. See CONTRIBUTING.md for setup and SECURITY.md for reporting vulnerabilities privately.

License

Code is MIT, textbook content is CC BY-SA 4.0 — both covered in LICENSE.

---

AI Safety Atlas is maintained by Markov Grey and the French Center for AI Safety.