AI safety newsletters cover alignment research, evaluations, governance and the debate over how fast frontier models should be deployed. The field is small enough that a handful of writers set the agenda, and several of them publish here.
The writing ranges from technical research digests to policy analysis. If you are new to the topic, start with one that explains the vocabulary before taking a side.
15 newsletters below, ranked by editorial quality and how recently they published.
Essays from Princeton's AI researchers (including Arvind Narayanan) arguing that AI is a transformative but normal technology rather than a path to superintelligence. Recent issues cover AI agent evaluation methods, whether AI should aim at augmenting rather than automating human labor, and debates over AI safety and existential risk.
Long-form essays on frontier AI, covering interpretability, AI safety scenarios, and how free societies might handle concentration of power. The author draws on political theory, history and literature, and has written from inside an AI lab.
Essays on AI agents, frontier lab economics and agent governance, mixing original simulation experiments with analysis and occasional fiction. Readers get a thoughtful, idiosyncratic view of how AI systems behave and how institutions might respond.
Research nonprofit blog on AI timelines forecasting, AGI and superintelligence risk, and policy proposals for managing frontier AI development. It publishes forecasting model updates, regulatory proposals, and testimony on AI safety.
Melanie Mitchell writes essays on developments in AI, examining claims about AI capabilities, risks and the nature of machine intelligence. Readers get critical, informed commentary that challenges hype and anthropomorphic framing.
Helen Toner's Substack on navigating a world with very advanced AI, covering the vagueness of AGI definitions, AI jaggedness, and the risks of personalized AI. Readers get careful, expert-informed argument about AI capabilities and governance, with occasional guest essays.
Zvi Mowshowitz's newsletter mixes fast weekly AI updates with longer world-model essays covering AI policy, interpretability, rationality, medicine and more. Readers get dense, opinionated commentary that digests policy fights and research claims in detail.
This newsletter offers annotated translations of Chinese-language articles on AI safety and governance, alongside original essays on US-China AI coordination. Readers get a view into how Chinese commentators engage with Silicon Valley debates on frontier AI risk.
A weekly AI and machine learning newsletter that pairs a thematic editorial with deeper explainers on new models, agents, world models and AI safety. Readers get framed analysis of the week's research and industry news, such as open-weight decision models, embedding models, self-improving agents and open-source strategy.
A weekly product management newsletter that examines how software is bought, how AI agents change buying and selling, and the safety of AI agents. Readers get analysis of product and go-to-market trends with concrete data from recent public purchases and research.
A weekly roundup of AI news and commentary, including text summaries and a podcast, covering frontier model releases, safety incidents, and AI mathematics. Readers get a digest of recent model launches, research findings, and editorial analysis of the AI industry.
A long-form newsletter by a writer covering AI and Silicon Valley culture through an anthropological lens, with reporting on the political backlash against AI, data centers, and AI safety. Readers get reported pieces, interviews with politicians and activists, and commentary on how AI is entering mainstream politics.
High Capacity is a newsletter and podcast by Kyle Chan, a Brookings fellow, examining China's AI policy, technology industry and US-China relations. Issues mix long-form interviews with China policy experts and commentary on AI safety, regulation and geopolitics.
Demis Hassabis, the AI researcher, argues that artificial general intelligence is likely a few years away and could be as transformative as electricity or fire, bringing major gains in medicine, energy and materials. The newsletter also addresses the risks of frontier and increasingly agentic AI systems and the need for careful, responsible development.
ControlAI is an advocacy newsletter that reports on alleged rogue AI incidents, such as attacks on government sites, Wikipedia and Tencent cloud, drawing on company disclosures and independent researchers. It pairs this news with calls for readers to contact lawmakers about AI risk.
One short email a week with hand-picked newsletters from the index. Free, no spam, unsubscribe anytime.
