Garrison Lovely argues in MIT Technology Review that AI safety advocates stay committed

Garrison Lovely argues in MIT Technology Review that AI safety advocates stay committed

Garrison Lovely’s article was published December 15, 2025. Lovely reports that AI safety advocates interviewed remain committed to their cause despite disappointment in GPT-5 and a political shift toward AI acceleration. They see debate about an AI bubble and GPT-5’s reception as distractions from their concern that advanced AI could become dangerous and difficult to control. They generally favor stronger regulation and worry that policymakers may treat recent setbacks as a reason not to prepare. The interviews convey a tension: some advocates welcome more time to understand AI systems, while worrying that policy and international coordination take time. Several say their estimates for dangerous systems have lengthened, and that their case does not depend on AGI arriving imminently. Their concern is that waiting for a clearly dangerous system could leave too little time to respond. The interviewees differ on timelines, how to read GPT-5 and the pace of progress; some also warn that aggressive predictions could damage AI safety’s credibility. Lovely reports their bottom line: whatever the timing, they believe the world is not ready for AGI.

Publication: d3634d68-6f9f-7af1-62c2-30c136b43bac

attribution — pass: The reading attributes the piece to Garrison Lovely in MIT Technology Review, which matches the byline and publisher. Interviewees' views stay attributed to them, not to EM. The title's "argues" is a slight overstatement for what is a reported interview piece. The body correctly uses "reports", and changing the title to "reports" would be the smallest repair.

Garrison Lovely is a freelance journalist and the author of Obsolete

summary — pass: The reading faithfully conveys the article's central finding: interviewees remain committed despite GPT-5 disappointment and the accelerationist shift. It also captures their more-time-but-worse-policy tension, the claim that the case is not about imminence, the credibility worry and the not-ready bottom line. AGI is defined beside the reading. The question follows from the coordination-time argument. Vivid voices such as Russell's asteroid analogy are omitted, but nothing essential is erased.

rather than feeling chastened or lost in the wilderness, they’re still deeply committed to their cause

date — pass: December 15, 2025 matches the UTC day of originalPublished (2025-12-15T10:00:00Z). The evidence is a publisher page, not an archive copy, so no archive disclosure is needed. The page was edited after publication (modified 2026-01-05), but the update note limits the edit to Kokotajlo's policy views, which the reading does not use. Later related-story sidebars are not relied on, and there is no hindsight framing.

This story has been updated to clarify some of Kokotajlo's views on AI policy.

support — pass: Each claim's excerpt is a verbatim match and supports it. The cited passages come from Lovely's framing and Krueger's interview, not the Kokotajlo material flagged in the update note, so they are plainly part of the original piece. Policy vulnerability, lengthened timelines, the coordination urgency and the bottom line are all stated in the source. No unsupported benefits, costs or forecasts are added.

worry that progress on policy—the implementation of the EU AI Act; the passage of the first major American AI safety bill, California’s SB 53

media — pass: No media was selected, and no verified media candidates were supplied. No media identity or relevance issue arises.