Orienting To 1 Year Agi Timelines

Open original

Nikola Jurkovic argues that superintelligence could arrive by early 2028 without slowdowns, so nontechnical and technical safety people should prioritize aggressive pacing or a pause within the next 6-12 months. He lays out which advocacy and technical work he thinks can actually pay off on that timeline.

  • METR's assessment of AI R&D automation at Anthropic is cited as strong evidence that timelines to superintelligence are now under two years, with a median around early 2028.
  • If AI dramatically automates AI research in 2027, avoiding catastrophe in many scenarios requires pacing or pausing development within 6-12 months.
  • Most nontechnical safety work is judged unlikely to pay off on that horizon; the recommended routes are lab advocacy, California lobbying, and federal lobbying after a major incident.
  • The bottleneck, per the author, is political will rather than policy ideas.

The post, by Nikola Jurkovic and published on LessWrong in November 2026, argues that timelines to superintelligence have collapsed. Citing a METR assessment of AI R&D automation inside Anthropic, plus a tweet from Ryan Greenblatt extrapolating that work, the author puts the median for superintelligence without any slowdown in early 2028 and expects AI to substantially automate and accelerate its own research during 2027. From that premise the argument is that, in a meaningful share of worlds where the leading lab fails to solve alignment during an intelligence explosion, avoiding catastrophe would require aggressive pacing or a full pause within the next 6 to 12 months. The author is blunt that most nontechnical safety people are working on theories of change that won't pay off on that timescale. The recommended nontechnical strategies are three. Lab advocacy aims to get labs to pace or pause voluntarily, without government involvement. California lobbying targets state government for a law or executive order imposing a slowdown within a year. Federal lobbying is conditional: if an incident roughly an order of magnitude worse than the Hugging Face episode happens in the next six months, push Congress or the Trump administration to respond with pacing or a pause. The post treats the Hugging Face incident and METR's Anthropic assessment as the two pieces of evidence that moved people's risk estimates in 2026, and that observation underpins much of the technical advice. On the technical side, the author thinks the best work falls into a few buckets. Third parties should lay the groundwork for a pause by building assessments of AI progress, AI R&D automation, and compute allocation, and should establish evidence of risk through embedded assessments of incidents, misalignment, loss of control, and covert capabilities. Embedded means working closely with labs, which the author argues is how the key 2026 findings were produced. Lab employees are urged to cooperate fully with that work by granting access and staff time. Alongside this, the post says some researchers should still work on alignment and control of early automated researchers, since failing there could be fatal if superintelligence arrives quickly. The author's closing diagnosis is that the bottleneck is political will rather than a shortage of policy ideas.

Ask about this article