At BlueDot Impact, we’ve been widely surveying the literature for existing AI safety plans. This document contains our notes on the most relevant articles, after reading almost everything we could find that seemed potentially relevant (for a more complete list of everything we’ve considered, see here‣). Our overall conclusion is that unfortunately there is no existing public strategy that meets our ‣.

The notes within documents are summaries/hot takes, and not necessarily ordered/prioritised by importance.

Public blog version: Key paths, plans and strategies to AI safety success

Most useful to read

• AI governance needs a theory of victory
• Planning for Extreme AI Risks
• [private plan]
• Analysis of Global AI Governance Strategies
• What success looks like
• Catastrophic risks from unsafe AI (particularly from 23:35 onwards)
• A Narrow Path: How to Secure Our Future

Maybe useful to read

• Scenario planning for AI x-risk
• A Playbook for AI Risk Reduction (focused on misaligned AI)
• The Checklist: What Succeeding at AI Safety Will Involve - Bowman (2024)
• Survey on intermediate goals in AI governance