Frontier Risk Report

METR's May 2026 Frontier Risk Report. Both papers quote its statement that no agents had yet been observed autonomously earning or stealing enough to support their own inference, and cite its task-horizon trend, Redwood's money-making runs that earned nothing, and its assessment that internal agents plausibly had the means, motive, and opportunity for small rogue deployments. These fix the status of the thresholds and motivate prediction 1.

Cited in

both papers (Flagship: Where agent lineages stand, The threshold, Fig. 2, Fig. 3; Companion §1, §3, §5, §6.4, Fig. 2, Fig. 3, Fig. 4)

Used for

Prediction 1P7Trophic independence; thresholds (Fig. 2); prediction 1; Fig. 3a; P7

Strand

AI safety, alignment & autonomy evaluations · Autonomy, replication & dangerous-capability evaluations

Cite

@techreport{metr2026frontier,
  title = {Frontier Risk Report},
  author = {METR},
  year = {2026},
  howpublished = {METR},
  url = {https://metr.org/blog/2026-05-19-frontier-risk-report/},
}