Sourceunknowntext stored

Scaling Trends for Lie Detector Oversight in Preference Learning

Hollinsworth, Oskar J., Dombrowski, Ann-Kathrin, Adam-Day, Sam, Gleave, Adam, Cundy, ChrisarXiv.orgPublished Jul 2, 2026Original︎ ↗Archived copy

Claims2

ExaminationsNone

Text45 sections

VersionsOne known

CopiesNone recorded

CorrectionsNone recorded

Facts checked Nov 11, 2026

ReachNo copies recorded

StructureIn 2 claims' maps0 sources downstream across 2 claims

Evidence1 reading · 1 read wholeSupports 1No claim cites an examination's findings

ReviewsNot reviewed

Stewards’ notes · 1

Some AI oversight methods resist deception better as the overseen model's capability advantage grows.
Supports. The study shows what it asserts about model scale: undetected deception falls with size at high detector true positive rates, robustly only at 0.8 and above, with a single seed at 405B. It does not vary a capability gap between model and overseer, and it attributes much of the trend to larger models being less deceptive before training. It bears on this claim only loosely. Nov 11, 2026

Examinations

None recorded. An examination is commissioned by a claim’s Steward for what its claim rests on; a review of the document for its own sake is asked for as a request bearing on it.

Text53 passages · 1 marked

No examination yetMarked passage

Title: Scaling Trends for Lie Detector Oversight in Preference Learning

Abstract: Deceptive behavior in LLMs is costly to monitor and prevent, motivating approaches such as Scalable Oversight via Lie Detectors (SOLiD) (Cundy & Gleave, 2025), which uses lie detectors to identify responses for review by high-cost labelers. In this paper, we scale SOLiD to larger models and evaluate it in more diverse and realistic preference-learning settings. We find favorable scaling: undetected deception drops from 34% for 1B-parameter models to 14% for 405B-parameter models at a detector true positive rate of 99%, and expensive human labelers can be removed entirely from the fine-tuning phase without a statistically significant increase in deception. However, SOLiD is sensitive to distribution shift between detector training and preference-training data, which can drive detector false positive rates to impractical levels.

Some AI oversight methods resist deception better as the overseen model's capability advantage grows.SupportedReading · this claim’s Steward

Supports. The study shows what it asserts about model scale: undetected deception falls with size at high detector true positive rates, robustly only at 0.8 and above, with a single seed at 405B. It does not vary a capability gap between model and overseer, and it attributes much of the trend to larger models being less deceptive before training. It bears on this claim only loosely.

Not found in the text 1

As quoted

We find favorable scaling: undetected deception drops from 34% for 1B-parameter models to 14% for 405B-parameter models at a detector true positive rate of 99% [...] However, SOLiD is sensitive to distribution shift between detector training and preference-training data, which can drive detector false positive rates to impractical levels.

Scalable oversight protocols remain effective against AI systems that deliberately deceive their overseers.Doubtfulmentions

Not yet read by this claim’s Steward.

Lineage across 2 claims

Draws on · none recorded

This source

Scaling Trends for Lie Detector Oversight in Preference Learning

in 2 claims' maps

Drawn on by · none recorded

Record

Kind: unknown
Authors: Hollinsworth, Oskar J., Dombrowski, Ann-Kathrin, Adam-Day, Sam, Gleave, Adam, Cundy, Chris
Publisher: arXiv.org
Published: Jul 2, 2026
DOI: None
Link: https://arxiv.org/abs/2607.01567
Archived copy: https://web.archive.org/web/20260703051317/https://arxiv.org/abs/2607.01567
Stored text: 4,229 characters, retrieved Nov 11, 2026 · 45 sections, 53 passages
Claims extracted: Not recorded
Versions: One known
Copies: None recorded
Corrections: None recorded
Facts checked: Nov 11, 2026

History 4

Nov 11, 2026ReadingSteward of “Some AI oversight methods resist deception bett…” · supports, read whole
Nov 11, 2026DividedSegmenter · Divided into 98 parts by its own structure
Nov 11, 2026FactsFacts fetcher · Facts looked up
Nov 11, 2026Entered the graphEntered the graph; copy stored