What evidence would reveal that recursive acceleration or transformative scientific automation had begun?
If recursion begins inside frontier labs, external indicators may lag badly. Scenario exercises try to specify what observable early signals — hiring, publication patterns, capability jumps — would look like.
View on the map → · Open in Browse →
The detection instruments now exist: METR's frontier risk report — with access inside four major labs — finds heavy AI use in R&D workflows but no measurable compression of research timelines yet, and FutureSearch's one-year check of AI 2027 finds coding and hacking ahead of schedule, prompting a one-year pull-forward of the superhuman-coder milestone.
no dramatic speed-ups in the overall pace of progress attributed to AI R&D automation
METR's pilot of assessing AI agents used inside Anthropic, Google, Meta and OpenAI — an external instrument aimed exactly at detecting internal acceleration; heavy AI use in R&D workflows, no measurable compression of research timelines yet.
Parts of a 2025 forecast of dangerous AI have started to come true
Scores which AI 2027 indicators have materialized one year in: autonomous coding and hacking ahead of schedule, R&D bottlenecks less binding than forecast — a worked example of reading early acceleration signals.