Import AI 458: Reckoning with the future; and a singularity story

Details

In Import AI issue 458, Jack Clark—co-founder of Anthropic—publishes the text of his 2026 Cosmos HAI Lab Lecture delivered at Oxford's Institute for Ethics in AI, alongside a piece of fiction imagining a positive singularity. The essay is organized in three parts: a historical view of AI progress via a benchmark graph, personal and institutional lessons from Anthropic's experience, and a set of choices that individuals and societies now face. Clark's central argument is that AI systems are advancing faster than most people expect, and that when they become sufficiently powerful they will not resemble a normal technology. He warns against what he calls 'retreating from the present'—dismissing or ignoring the implications of continued AI progress—and argues instead for proactive engagement with the choices AI's advance creates: how to shape it, who benefits, and how to direct it. The essay is notable for its framing: Clark is not making technical claims but a philosophical and social argument from inside one of the frontier AI labs. The fictional story complements the essay by attempting to imagine what optimistic long-term outcomes might look like. For practitioners, the primary value is a window into how a key figure at a frontier lab thinks about the macro trajectory—not a policy proposal or product announcement, but a directional statement about where AI is headed and why we should plan for its success rather than its failure.