From Mexico — I’m skipping the week-old GPT-6 Astra rehash this morning. The fresher wire is governance. OpenAI Foundation announced Paul Christiano is joining its board and the Safety and Security Committee (SSC), chaired by Zico Kolter. He’ll also sit as a non-voting observer on the OpenAI Group PBC board. Same day, Christiano’s own Substack dropped a personal statement that doesn’t soft-pedal the stakes.
Per that statement: “there is a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term.” Then the line that traveled overnight — he does not think “the AI industry in general, including OpenAI, is currently on track to reduce this risk to an acceptable level.” He’s joining because he thinks OpenAI could cut risk if it rises to the occasion. Business Insider and The Guardian both carried the warning into Sep 10 coverage.
Who he is — and what the SSC seat means
The Foundation post sketches the résumé: Senior Tech Advisor at CAISI (NIST / Commerce), founder of the Alignment Research Center, and OpenAI alignment lead from 2017–2021 (including early RLHF work). Bret Taylor’s quote there frames him as someone who will challenge assumptions on the board. Christiano’s own line on the appointment: AI capabilities moved fast last year, alignment is still hard, and the SSC’s job just got heavier. Footnote on the Foundation page: as a CAISI adviser he recuses from OpenAI-related government matters and model evaluations.
The numbers — and the “most people could die” line
I’m quoting the Substack directly so nobody has to trust a paraphrase. On quantifying uncertainty: “an all-things-considered risk of 4% over the next year and 15% over the next three years” — subjective beliefs, not a precise model. On the failure mode: “If we build superintelligence without more robust alignment I expect we will permanently lose control of it. If that happens then most people could die.” He also flags automated AI R&D as a path to a rapid intelligence explosion, and says public evidence from recent incidents makes misaligned reward-seeking more than a theory. He wants domestic and international coordination — and says frontier labs can still unilaterally raise mitigations, share evidence, and push shared standards.
Context, not the lead: this lands in the same week as Anthropic’s Coxon resignation chatter (I already covered that yesterday). Different story. This morning’s post is the OpenAI Foundation appointment plus Christiano’s own risk math. Sources: Christiano Substack, OpenAI Foundation, Business Insider, The Guardian.
Hero image: empty modern office hallway by Nastuh Abootalebi on Unsplash (Unsplash License). Cropped, graded, and lightly grained by Tech & AI Pulse. Face-free still — no people, no logos.
