AI Safety Project

Project site: https://dunchead.github.io/ai-safety/

The AI Safety Project aims to map landscape of AI risk and understand the challenges we must overcome as a global community.

The field of AI development is moving at such a pace that there's a real chance we may fall victim to risks before we fully understand them, or are even aware of them. This is probably already happening.

Community-driven initiatives are essential to keep up with this rapidly evolving landscape. The goals of this project are:

• Create a simple framework for understanding the landscape of AI risk and the main challenges we need to overcome
• Share key resources and potential solutions
• Provide a space for community collaboration that is not driven by economic incentives

Please contribute!

This is a collaborative project and a work in progress. We encourage contributions for any of the following:

• Additional risks/challenges
• Suggested solutions
• Key links/resources

Feel free to submit your own ideas.

Bugfixes/improvements for the website itself are also welcome.

Contributions can be made in two ways:

1. As posts in the discussion forum under 'AI risks, challenges and solutions'
2. As pull requests (most content is in script.js)

Run locally

This is a static site, so no build step is required. To run it with Python's simple static server, serve the parent directory and open the project path:

cd ..
python3 -m http.server 8000

Then visit:

• http://localhost:8000/ai-safety/
• http://localhost:8000/ai-safety/risks.html
• http://localhost:8000/ai-safety/challenges.html

The public site uses clean URLs such as /ai-safety/risks and /ai-safety/challenges. Those are handled by .htaccess, which maps clean URLs to the corresponding .html files and redirects direct .html requests back to clean URLs. Python's simple server does not process .htaccess, so use the .html URLs when testing locally that way.

Contributor guidelines

This is a high-level map, not an encyclopedia.

The goal is to create a clear outline of the major issues with links to only the most useful resources for further reading; we're not trying to compile a detailed or comprehensive list. Too much information is part of the problem we're trying to solve.

Yes please:
✅ Key points not already covered, concisely written
✅ Links/resources that are especially clear, original or significant
✅ Restructuring that improves the content while maintaining the clarity of design

No thanks:
❌ Verbose or repetitive text
❌ Links/resources that do not offer much additional value
❌ Non-essential restructuring of the design/content