WhitzardNUWA Lab ↗
NUWANUWA INDEX

Search NUWA Index

Research · Benchmarks · Findings · Perspectives

中文
← Library

Risk Evaluation · Study

Auditing Games for Sandbagging

NUWA research note · independent review pending

First published: Source checked: 2026-10-25
Risk monitoring
Original source ↗Cite this record ↓Deception →
NUWA Research NoteResearch questionDesign and assumptionsSource findingsNUWA interpretationLimits and contrary evidenceResearch and decision relevanceInstitutions and rolesOriginal resources

Model-assisted reading · Selected sections of the original full text or official article; a complete section-by-section audit remains pending.

At a glance

NUWA interpretation: elicitation uplift alone is not diagnostic of intentional capability hiding.

Research question

Can auditors distinguish trained sandbaggers from genuinely weaker models?

Design and assumptions

A red team fine-tunes five model organisms; a blue team applies black-box, activation-probe and training-based detection under explicit game constraints.

Source findings

Black-box approaches did not reliably discriminate sandbaggers.

One correct training demonstration could elicit full performance, but benign models also improved, creating false positives.

NUWA interpretation

NUWA interpretation: elicitation uplift alone is not diagnostic of intentional capability hiding.

Limits and contrary evidence

Trained organisms are proxies, and detector robustness depends on the red-team construction and access regime.

Research and decision relevance

Model recommendation: include benign weak-model controls and adversarially constructed organisms when validating sandbagging detection.

Precise source locations
  • Auditing Games for Sandbagging · AbstractSelected original resource https://arxiv.org/html/2512.07810; sha256:0c00d4cce9c41437666e2e086b86648e87bcb735c9624e4bc87c85eb9fc63470; archived source capture; exact capture time not independently established
  • Auditing Games for Sandbagging · §3 Game SetupSelected original resource https://arxiv.org/html/2512.07810; sha256:0c00d4cce9c41437666e2e086b86648e87bcb735c9624e4bc87c85eb9fc63470; archived source capture; exact capture time not independently established
  • Auditing Games for Sandbagging · §6 Blue Team – Detection ResultsSelected original resource https://arxiv.org/html/2512.07810; sha256:0c00d4cce9c41437666e2e086b86648e87bcb735c9624e4bc87c85eb9fc63470; archived source capture; exact capture time not independently established
  • Auditing Games for Sandbagging · §7 DiscussionSelected original resource https://arxiv.org/html/2512.07810; sha256:0c00d4cce9c41437666e2e086b86648e87bcb735c9624e4bc87c85eb9fc63470; archived source capture; exact capture time not independently established

Source checking, extraction, editorial approval and replication are recorded separately. Methodology →

Institutions and roles

Research by UK AI Security InstituteFAR.AIAnthropic

Authors (13)

Jordan Taylor · Sid Black · Dillon Bowen · Thomas Read · Satvik Golechha · Alex Zelenka-Martin · Oliver Makins · Connor Kissane · Kola Ayonrinde · Jacob Merizian · Samuel Marks · Chris Cundy · Joseph Bloom

Cite and reuse

Cite the original publication for its findings. Cite this NUWA record separately for its explanation and retained snapshot.

BibTeX ↓RIS ↓Record + source metadata · JSON ↓Reading coverage · JSONSuggest a correction
Original publication · BibTeX
@misc{auditing-games,
  title = {Auditing Games for Sandbagging},
  url = {https://arxiv.org/abs/2512.07810},
  author = {Jordan Taylor and Sid Black and Dillon Bowen and Thomas Read and Satvik Golechha and Alex Zelenka-Martin and Oliver Makins and Connor Kissane and Kola Ayonrinde and Jacob Merizian and Samuel Marks and Chris Cundy and Joseph Bloom},
  year = {2025},
  date = {2025-12-08},
  version = {arXiv:2512.07810v1},
  note = {arXiv:2512.07810v1}
}
NUWA record · BibTeX
@misc{auditing-games-nuwa,
  title = {Auditing Games for Sandbagging — NUWA research record},
  url = {https://whitzard.tech/en/nuwa/whitzard-index/work/auditing-games/},
  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--auditing-games.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--auditing-games.json.}
}

Evidence snapshot 2026-10-29.1 · Index updates · RSS

References

  1. [1] Jordan Taylor, Sid Black Dillon Bowen Thomas Read Satvik Golechha, et al.. Auditing Games for Sandbagging. 2025-12-08. v1 [arXiv]; Author affiliations on the original paper title page; Abstract; §3 Game Setup