AI Navigate

AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors

arXiv cs.CL / 3/11/2026

Ideas & Deep AnalysisTools & Practical UsageModels & Research

Key Points

  • AuditBench is a newly introduced alignment auditing benchmark consisting of 56 language models implanted with 14 different hidden concerning behaviors that the models do not confess to when asked.
  • The models exhibit varying levels of subtlety and overt behavior, and different training methods are used both to implant the behaviors and to train models not to confess.
  • An investigator agent is developed to autonomously use a configurable set of auditing tools, allowing evaluation of tool efficacy in the context of alignment auditing.
  • Results highlight a tool-to-agent gap where tools that perform well individually do not necessarily improve auditing performance when employed by the investigator agent.
  • Audit effectiveness depends on training methods, with models trained on synthetic documents easier to audit than those trained on demonstrations, and adversarial training making auditing more difficult; the benchmark and tools are released for future research advancement.

Computer Science > Computation and Language

arXiv:2602.22755 (cs)
[Submitted on 26 Feb 2026 (v1), last revised 9 Mar 2026 (this version, v3)]

Title:AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors

View a PDF of the paper titled AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors, by Abhay Sheshadri and 7 other authors
View PDF HTML (experimental)
Abstract:We introduce AuditBench, an alignment auditing benchmark. AuditBench consists of 56 language models with implanted hidden behaviors. Each model has one of 14 concerning behaviors--such as sycophantic deference, opposition to AI regulation, or secret geopolitical loyalties--which it does not confess to when directly asked. AuditBench models are highly diverse--some are subtle, while others are overt, and we use varying training techniques both for implanting behaviors and training models not to confess. To demonstrate AuditBench's utility, we develop an investigator agent that autonomously employs a configurable set of auditing tools. By measuring investigator agent success using different tools, we can evaluate their efficacy. Notably, we observe a tool-to-agent gap, where tools that perform well in standalone non-agentic evaluations fail to translate into improved performance when used with our investigator agent. We find that our most effective tools involve scaffolded calls to auxiliary models that generate diverse prompts for the target. White-box interpretability tools can be helpful, but the agent performs best with black-box tools. We also find that audit success varies greatly across training techniques: models trained on synthetic documents are easier to audit than models trained on demonstrations, with better adversarial training further increasing auditing difficulty. We release our models, agent, and evaluation framework to support future quantitative, iterative science on alignment auditing.
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2602.22755 [cs.CL]
  (or arXiv:2602.22755v3 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2602.22755
Focus to learn more
arXiv-issued DOI via DataCite

Submission history

From: Abhay Sheshadri [view email]
[v1] Thu, 26 Feb 2026 08:43:07 UTC (1,303 KB)
[v2] Tue, 3 Mar 2026 06:42:39 UTC (1,090 KB)
[v3] Mon, 9 Mar 2026 18:35:46 UTC (1,089 KB)
Full-text links:

Access Paper:

    View a PDF of the paper titled AuditBench: Evaluating Alignment Auditing Techniques on Models with Hidden Behaviors, by Abhay Sheshadri and 7 other authors
  • View PDF
  • HTML (experimental)
  • TeX Source
Current browse context:
cs.CL
< prev   |   next >
Change to browse by:
cs

References & Citations

export BibTeX citation Loading...

BibTeX formatted citation

×
Data provided by:

Bookmark

BibSonomy logo Reddit logo
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?)
Links to Code Toggle
Papers with Code (What is Papers with Code?)
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?)
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.