A menu of readings for building context in AI safety beyond your own project. Nothing here is required. If you'd prefer a structured course with a cohort, we recommend BlueDot Impact's free online AI safety courses. The FAQ in the SPAR Mentee Guide (Fall ‘26) explains how to apply as a SPAR mentee.
<aside>
🔐
Core Reading List
If you only read a few things, start here.
Risks from AI:
- https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities
- https://www.lesswrong.com/posts/HBxe6wdjxK239zajf/what-failure-looks-like
- The Superintelligent Will
AI safety research:
- https://www.anthropic.com/research/alignment-faking (Anthropic and Redwood Research)
- https://blog.redwoodresearch.org/p/the-case-for-ensuring-that-powerful (Redwood Research)
- https://www.lesswrong.com/posts/dZFpEdKyb9Bf4xYn7/tips-for-empirical-alignment-research (Ethan Perez)
AI strategy:
- https://www.lesswrong.com/posts/E8n93nnEaFeXTbHn5/plans-a-b-c-and-d-for-misalignment-risk
- https://www.lesswrong.com/posts/vjAM7F8vMZS7oRrrh/how-do-we-more-safely-defer-to-ais
- https://www.forethought.org/research/short-timelines-arent-obviously-higher-leverage
AI forecasting:
- https://www.dwarkesh.com/p/carl-shulman
- https://asteriskmag.com/issues/10/does-ai-progress-have-a-speed-limit
- https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/ (METR)
Generalist work:
- https://usefulfictions.substack.com/p/how-to-be-more-agentic
- https://80000hours.org/2026/01/the-most-overlooked-roles-in-ai-safety/ (80,000 Hours)
- https://80000hours.org/podcast/episodes/tara-mac-aulay-operations-mindset/ (Tara Mac Aulay)
</aside>
<aside>
🧱
General AI safety context
General resources:
- https://www.youtube.com/c/robertmilesai
- https://www.cold-takes.com/most-important-century/
- https://80000hours.org/problem-profiles/
- https://en.wikipedia.org/wiki/Human_Compatible
- https://www.alignmentforum.org/library
- https://www.lesswrong.com/tag/ai
- https://forum.effectivealtruism.org/topics/ai-safety
- https://80000hours.org/podcast/
- https://www.lesswrong.com/posts/HzKuzrKfaDJvQqmjh/responsible-scaling-policy-v3
- https://metr.org/blog/2026-05-19-frontier-risk-report/ (METR)
- https://internationalaisafetyreport.org/publication/international-ai-safety-report-2026
Risks from AI:
- https://www.lesswrong.com/posts/uMQ3cqWDPHhjtiesc/agi-ruin-a-list-of-lethalities
- https://www.lesswrong.com/posts/HBxe6wdjxK239zajf/what-failure-looks-like
- The Superintelligent Will
- https://ai-2027.com/ (AI Futures Project)
- https://gradual-disempowerment.ai/
- https://www.youtube.com/watch?v=5KVDDfAkRgc (AI in Context video)
- https://www.youtube.com/watch?v=KUieFuV1fuo
- https://www.youtube.com/watch?v=GT_sXIUJPUo
- https://www.forethought.org/research/ai-enabled-coups-how-a-small-group-could-use-ai-to-seize-power
- https://www.cold-takes.com/ai-could-defeat-all-of-us-combined/
- https://www.cold-takes.com/how-we-could-stumble-into-ai-catastrophe/
- https://www.safe.ai/ai-risk
- https://arxiv.org/abs/2306.12001
- https://arxiv.org/abs/2206.13353
</aside>
<aside>
🔬
Technical AI safety
If your project is technical, your mentor's suggested readings come first. These build context on the main empirical safety agendas.
Alignment failures:
- https://www.deepmind.com/blog/specification-gaming-the-flip-side-of-ai-ingenuity (Google DeepMind)
- https://youtu.be/bJLcIBixGj8 (Robert Miles video)
- https://www.cold-takes.com/why-ai-alignment-could-be-hard-with-modern-deep-learning/ (Ajeya Cotra)
- https://arxiv.org/abs/2209.00626
- https://www.anthropic.com/research/alignment-faking (Anthropic and Redwood Research)
- https://www.anthropic.com/research/sleeper-agents-training-deceptive-llms-that-persist-through-safety-training (Anthropic)
- https://www.anthropic.com/research/emergent-misalignment-reward-hacking (Anthropic)
Control and oversight:
- https://blog.redwoodresearch.org/p/the-case-for-ensuring-that-powerful (Redwood Research)
- https://arxiv.org/abs/2312.06942
- Weak-to-strong generalization (OpenAI)
Interpretability and evals:
- https://www.anthropic.com/research/tracing-thoughts-language-model (Anthropic)
- https://transformer-circuits.pub/2023/interpretability-dreams/index.html (Chris Olah)
- https://www.anthropic.com/research/auditing-hidden-objectives (Anthropic)
Doing research well:
- https://www.lesswrong.com/posts/dZFpEdKyb9Bf4xYn7/tips-for-empirical-alignment-research (Ethan Perez)
- https://www.alignmentforum.org/posts/hjMy4ZxS5ogA9cTYK/how-i-think-about-my-research-process-explore-understand (Neel Nanda)
- http://joschu.net/blog/opinionated-guide-ml-research.html (John Schulman)
- Research as a Stochastic Decision Process (Jacob Steinhardt)
- https://www.alignmentforum.org/posts/EF5M6CmKRd6qZk27Z/my-research-methodology
- https://www.lesswrong.com/posts/eJGptPbbFPZGLpjsp/highly-opinionated-advice-on-how-to-write-ml-papers
- https://www.lesswrong.com/posts/37sHjeisS9uJufi4u/scholarship-how-to-do-it-efficiently
- https://www.lesswrong.com/w/research-taste
- https://www.neelnanda.io/mechanistic-interpretability/getting-started
- https://www.lesswrong.com/posts/SXJGSPeQWbACveJhs/the-best-tacit-knowledge-videos-on-every-subject
</aside>
<aside>
😈
AI strategy
General resources:
- https://blog.redwoodresearch.org/
- https://newsletter.forethought.org/
On current AI alignment:
- https://www.lesswrong.com/posts/WewsByywWNhX9rtwi/current-ais-seem-pretty-misaligned-to-me
- https://www.lesswrong.com/posts/epjuxGnSPof3GnMSL/alignment-remains-a-hard-unsolved-problem
Plans for the future:
- https://www.lesswrong.com/posts/vjAM7F8vMZS7oRrrh/how-do-we-more-safely-defer-to-ais
- https://www.lesswrong.com/posts/E8n93nnEaFeXTbHn5/plans-a-b-c-and-d-for-misalignment-risk
- https://www.forethought.org/research/short-timelines-arent-obviously-higher-leverage
- https://www.forethought.org/research/viatopia
</aside>
<aside>
📈
</aside>
<aside>
🥁
</aside>
<aside>
🦉
</aside>
<aside>
💰
</aside>
<aside>
🧩
</aside>