Skip to content
AI News HubLIVE
Original source2 min read

Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research

Summary

A new red-team/blue-team study finds that human expert auditors, despite having full research transcripts and the ability to run experiments, barely distinguish sabotaged from benign AI safety research (AUROC 0.59, only 17% full identification). Frontier models like Fable 5.1 outperform humans even without experiment access (AUROC 0.73, 32% full identification). Identification drops further when the saboteur strategically chooses which experiments to attack. The results suggest current auditing methods are ill-equipped to detect research sabotage by expert-level AI agents.

SourcearXiv AIAuthor: Keshav Shenoy, Anthony Zang, Eric Gan, Jacqueline Tay, Amy Ngo, Sebastian Prasanna, Julian Stastny, Vivek Hebbar
Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research
Report an error

The correction channel is not available yet. You can copy the article reference below for later.

Correction instructions
Read article

[Submitted on 16 Nov 2026]

Title:Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research

View a PDF of the paper titled Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research, by Keshav Shenoy and 7 other authors

View PDF HTML (experimental)

Abstract:As AI systems are increasingly used to automate AI R&D, misaligned systems could sabotage research intended to improve model safety and alignment. To stress-test defenses against such sabotage, we use a human expert as a proxy saboteur on AI safety research. We conduct a red-team/blue-team game where a human researcher runs AI safety experiments from scratch and auditors try to tell which were sabotaged and how. Given access to the full research transcript and the ability to run experiments, human auditors mostly fail to distinguish sabotaged from benign research: they achieve an AUROC of 0.59, not significantly better than chance, and fully identify the specific sabotage in only 17% of cases. Despite not being allowed to run experiments, the best frontier models outperform human auditors (Fable 5.1: AUROC 0.73, 32% full identification) within noise and weaker models perform comparably. When the saboteur chooses which experiments are attacked, rather than being required to sabotage every question given, sabotage identification drops further. Our results suggest that research sabotage by expert-level agents is difficult to detect with current auditing methods and provide a new setting for studying research sabotage.

Subjects:

Artificial Intelligence (cs.AI)

Cite as: arXiv:2611.21832 [cs.AI]

(or arXiv:2611.21832v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2611.21832

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Keshav Shenoy [view email] [v1] Mon, 16 Nov 2026 19:52:38 UTC (669 KB)

Full-text links:

Access Paper:

View a PDF of the paper titled Auditors Struggle to Catch Sabotage In End-to-End Transcripts of AI Safety Research, by Keshav Shenoy and 7 other authors

View PDF

HTML (experimental)

TeX Source

view license

Additional Features

Audio Summary

Current browse context:

cs.AI

new | recent | 2026-11

Change to browse by:

cs

References & Citations

SciX

Google Scholar

Semantic Scholar

Loading...

Data provided by:

Bibliographic Tools

Bibliographic and Citation Tools

Bibliographic Explorer Toggle

Bibliographic Explorer (What is the Explorer?)

Connected Papers Toggle

Connected Papers (What is Connected Papers?)

Litmaps Toggle

Litmaps (What is Litmaps?)

scite.ai Toggle

scite Smart Citations (What are Smart Citations?)

Code, Data, Media

Code, Data and Media Associated with this Article

alphaXiv Toggle

alphaXiv (What is alphaXiv?)

Links to Code Toggle

CatalyzeX Code Finder for Papers (What is CatalyzeX?)

DagsHub Toggle

DagsHub (What is DagsHub?)

GotitPub Toggle

Gotit.pub (What is GotitPub?)

Huggingface Toggle

Hugging Face (What is Huggingface?)

ScienceCast Toggle

ScienceCast (What is ScienceCast?)

Demos

Demos

Replicate Toggle

Replicate (What is Replicate?)

Spaces Toggle

Hugging Face Spaces (What is Spaces?)

Spaces Toggle

TXYZ.AI (What is TXYZ.AI?)

Related Papers

Recommenders and Search Tools

Link to Influence Flower

Influence Flower (What are Influence Flowers?)

Core recommender toggle

CORE Recommender (What is CORE?)

Author

Venue

Institution

Topic

About arXivLabs

arXivLabs: experimental projects with community collaborators

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)

Key points and analysis

Article intelligence

EngineersAdvanced

Key points

  • Human auditors achieve only AUROC 0.59 and 17% full identification, not significantly better than chance.
  • Frontier model Fable 5.1 outperforms humans with AUROC 0.73 and 32% full identification despite being barred from running experiments.
  • Sabotage detection becomes even harder when the saboteur selects which experiments to attack.
  • The study highlights the difficulty of detecting expert-level research sabotage with current auditing methods.

Highlights and analysis are generated automatically and may contain errors. Check the original source.