NUWA Risk Discovery · Study
Large language model-powered AI systems achieve self-replication with no human intervention
NUWA research note · independent review pending
NUWA Research Note
Research questionDesign and assumptionsSource findingsNUWA interpretationLimits and contrary evidenceResearch connectionsResearch and decision relevanceInstitutions and rolesOriginal resourcesModel-assisted reading · Selected sections of the original full text or official article; a complete section-by-section audit remains pending.
At a glance
The expanded study finds copying success across several model families, under an explicitly supplied replication task.
Research question
How broadly does end-to-end replication appear when a common scaffold is paired with different models?
Design and assumptions
A common scaffold is evaluated across a model panel with repeated trials in resettable containers; success requires a separated running copy with the same functionality.
Source findings
Success is not confined to the largest tested models.
The paper reports associations between general benchmark capability and copying success, and additional selected adaptation trajectories.
NUWA interpretation
NUWA interpretation: model breadth strengthens evidence that the mechanism is not unique to one checkpoint, without making access conditions irrelevant.
Limits and contrary evidence
Small per-model trial counts, benchmark correlations and selected trajectories do not identify a universal capability threshold or spontaneous propensity.
Research connections
NUWA connection: the earlier self-replication paper is a narrower precursor; the network-replication paper adds obtaining target access to a copying pipeline.
Research and decision relevance
NUWA recommendation: compare only deployments with matching target access and success criteria; retain precursor and expanded versions separately.
Precise source locations
- Large language model-powered AI systems achieve self-replication with no human intervention · 2 Evaluation on the self-replication capabilitySelected original resource https://arxiv.org/html/2503.17378; sha256:bdb7bb0349475c85f6e6ece36a2d956bc7071ae0f574d4d5125a7909c8e28c54; archived source capture; exact capture time not independently established
- Large language model-powered AI systems achieve self-replication with no human intervention · 3 MethodsSelected original resource https://arxiv.org/html/2503.17378; sha256:bdb7bb0349475c85f6e6ece36a2d956bc7071ae0f574d4d5125a7909c8e28c54; archived source capture; exact capture time not independently established
- Large language model-powered AI systems achieve self-replication with no human intervention · 4.1 Emergence of the self-replication capabilitySelected original resource https://arxiv.org/html/2503.17378; sha256:bdb7bb0349475c85f6e6ece36a2d956bc7071ae0f574d4d5125a7909c8e28c54; archived source capture; exact capture time not independently established
- Large language model-powered AI systems achieve self-replication with no human intervention · 6 DiscussionSelected original resource https://arxiv.org/html/2503.17378; sha256:bdb7bb0349475c85f6e6ece36a2d956bc7071ae0f574d4d5125a7909c8e28c54; archived source capture; exact capture time not independently established
Source checking, extraction, editorial approval and replication are recorded separately. Methodology →
Institutions and roles
Authors (6)
Xudong Pan · Jiarun Dai · Yihe Fan · Minyuan Luo · Changyi Li · Min Yang
Cite and reuse
Cite the original publication for its findings. Cite this NUWA record separately for its explanation and retained snapshot.
Original publication · BibTeX
@misc{self-replication-extended,
title = {Large language model-powered AI systems achieve self-replication with no human intervention},
url = {https://arxiv.org/abs/2503.17378},
author = {Xudong Pan and Jiarun Dai and Yihe Fan and Minyuan Luo and Changyi Li and Min Yang},
year = {2025},
date = {2025-03-14},
version = {arXiv:2503.17378v2},
note = {arXiv:2503.17378v2}
}
NUWA record · BibTeX
@misc{self-replication-extended-nuwa,
title = {Large language model-powered AI systems achieve self-replication with no human intervention — NUWA research record},
url = {https://whitzard.tech/en/nuwa/whitzard-index/work/self-replication-extended/},
version = {2026-10-29.1},
organization = {NUWA Frontier AI Safety Lab},
howpublished = {Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--self-replication-extended.json},
note = {NUWA research note. Source-reading scope and independent review status are stated on the record. Snapshot 2026-10-29.1. Retained note edition: https://whitzard.tech/nuwa/whitzard-index/data/record-editions/2026-10-29.1--self-replication-extended.json.}
}
Evidence snapshot 2026-10-29.1 · Index updates · RSS
References
- [1] Yihe Fan (范一禾), Minyuan Luo (罗铭源), et al.. Large language model-powered AI systems achieve self-replication with no human intervention. 2025-03-14. v2 [arXiv]; Author affiliations on the original paper title page; 2 Evaluation on the self-replication capability; 3 Methods