RepliBench: Evaluating the Autonomous Replication Capabilities of Language Model Agents

A preprint introducing RepliBench, an evaluation of language model agents' capabilities for autonomous replication. Both papers cite it in the thresholds figure for the finding that autonomous agents self-replicate only in the laboratory. It supports the status claim that, as of October 2026, no lineage has crossed feralization or naturalization.

Cited in

both papers (Flagship: Fig. 2; Companion Fig. 2)

Used for

Thresholds; Fig. 2

Strand

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

Cite

@misc{black2025replibench,
  title = {RepliBench: Evaluating the Autonomous Replication Capabilities of Language Model Agents},
  author = {Sid Black and Asa Cooper Stickland and Jake Pencharz and Oliver Sourbut and Michael Schmatz and Jay Bailey and Ollie Matthews and Ben Millwood and Alex Remedios and Alan Cooney},
  year = {2025},
  url = {https://arxiv.org/abs/2504.18565},
}