Import AI has been running since 2016 and it is written by Jack Clark, who was at OpenAI and is now a co-founder of Anthropic. That biography is both why the newsletter is as good as it is and the main thing to hold in mind while reading it. What distinguishes Import AI from every other AI newsletter is that it does the hard part. Anyone can summarise a paper.
Clark tells you why the result matters, what it implies about where capabilities are heading, what it means for policy, and how it connects to three other things from previous months. Each item has a section explaining the significance, and those sections are frequently more valuable than the paper summaries they follow. It is the difference between being informed and being oriented. The combination of research, policy and safety in one place is rare and genuinely useful.
Most sources pick one. Research newsletters cover papers, policy publications cover regulation, safety writing sits in its own community. Clark works across all three, and because AI policy is being written in response to research results, the connections he draws are the ones that actually matter. Someone trying to understand why a particular regulatory proposal exists will get more from Import AI than from reading the proposal.
The paper selection reaches further than the usual. Compute trends, evaluation methodology, surveillance applications, military use, work from Chinese labs that Western coverage frequently ignores, and results from smaller groups that never get a press release. Over years of reading, the picture you accumulate is broader than what the announcement driven newsletters produce. Now the thing that has to be said.
Clark is a co-founder of Anthropic and a prominent voice in AI policy. He is transparent about this and does not hide it. It still means that a person with a substantial commercial and reputational stake in how AI is regulated and how AI capabilities are perceived is providing your weekly analysis of AI capabilities and regulation. He is scrupulous in a way many would not be, and the framing is not neutral, because no one's is.
Read it alongside sources with different institutional positions. Academic critics, people sceptical of scaling, writers outside the lab ecosystem entirely. The barrier to entry is real. This assumes you know what a benchmark is, roughly how training works, what the current model landscape looks like, and enough policy vocabulary to follow a discussion of export controls or compute thresholds.
Someone new to AI will find it hard going and should build the foundation first. It is a newsletter for people already in the conversation. The weekly cadence means it is not a news source. Things you read about here you will often have seen days earlier.
The value is the interpretation, which needs the time. Read TLDR AI daily for coverage and Import AI weekly for what it meant. The speculative fiction that closes each issue is the most divisive element. Short pieces imagining futures shaped by the technology.
Some are genuinely affecting and some do not land, and readers seem to either look forward to them or skip past them entirely. I am mostly in the first camp and I recognise it is a matter of taste. Policy coverage is heavily weighted towards the US and UK, with the EU appearing regularly and the rest of the world less so. Understandable given where the author sits and a limitation if you need a global picture.
My four point three is for the most substantive analysis available in this format, sustained for the better part of a decade, for free. It is marked down for a perspective that comes from inside the industry it covers and for a level of assumed knowledge that makes it useless as an entry point. If you have the background, read it, and read something adversarial alongside it.