Import AI 459: AI oversight, scaling laws for protein folding models, and putting a price on risk

The 459th issue of the Import AI newsletter, which Jack Clark, a co-founder of Anthropic and its head of policy, has written since 2016, came out on June 1, 2026. The new issue combines three topics: the difficulty of overseeing artificial intelligence systems, scaling laws for models that predict protein folding, and attempts to put a monetary value on the risk of an existential catastrophe that AI systems may pose.

Three topics in issue 459

Import AI is one of the longest-running and most respected newsletters on AI policy and research. Since 2016, Jack Clark has used it to collect short reviews of the week's significant papers, reports and events, with his own commentary aimed at policymakers, researchers and anyone tracking where AI is heading beyond product announcements. Issue 459 continues that tradition, linking technical progress, in the form of scaling laws for narrow scientific models, with questions of AI safety and governance across the whole industry.

Jack Clark is a prominent figure in the debate over AI regulation. Besides his work at Anthropic, he has written Import AI for many years as a personal rather than a corporate project, which gives the newsletter a reputation as an independent source: it regularly raises topics that are uncomfortable for the industry as a whole, not just ones convenient for promoting particular products. That is why researchers, journalists and staff at organizations that track AI policy cite Import AI consistently. The issue number itself, 459 over years of regular publication, shows a consistency that is rare for a personal project of this scale and format.

Why AI oversight is getting harder

The difficulty of oversight reflects a broader discussion in the AI safety community. As models grow more powerful and more agentic, traditional control mechanisms (human review of outputs, pre-deployment testing, limits on access to tools) face a widening gap between the speed at which systems make decisions and the speed at which people can check them. The more autonomous a system becomes, the harder it is to build oversight that does not turn into either a bottleneck that cancels out the benefits of automation or a formality that offers no real protection.

Key facts:

• Import AI issue 459 was published on June 1, 2026
• The newsletter's author is Jack Clark, co-founder and head of policy at Anthropic
• Topics: AI oversight, scaling laws for protein folding models, the price of extinction risk
• Import AI has been published regularly since 2016

Why put a price on extinction risk, and what protein folding models offer

The issue's other two topics look far apart at first glance, but both show how the AI industry is moving in two directions at once: the practical and the existential. Models that predict protein folding are one of the most successful applications of deep learning in science. Gains in their accuracy and scale directly speed up drug development and the understanding of biological mechanisms, and scaling laws for such models show how predictably their capabilities grow with more data and compute, much as with language models.

Alongside this, the discussion of a "price" for extinction risk is an attempt to translate the abstract debate about AI's existential risks into a language economists and insurers understand. If a risk can be estimated in money, even roughly, it can be handled like any other systemic risk: built into regulation, insurance and corporate governance, rather than left solely to philosophical argument. Combining these three topics in one issue is typical of Import AI: the newsletter consistently shows that technical progress and AI safety do not develop in parallel but intertwine.

For regular readers, this feature of the format, which pairs concrete scientific results with systemic safety questions in a single issue, explains why Import AI remains one of the few sources that engineers and policymakers read with equal attention.