analiza aiqbench
This paper challenges the prevailing narrative that automated red-teaming renders human evaluators obsolete by exposing a critical "threat-model coverage gap" in current safety benchmarks. The authors argue that because benchmarks are fixed by developers, they inherently miss harms relevant to specific deployment contexts or non-English prompts, a limitation shared across fields like cryptography and clinical trials. Consequently, the study asserts that maintaining diverse human evaluators is methodologically necessary to ensure comprehensive threat coverage rather than just internal validation on predefined sets.
A Translational Note on AI Safety Evaluation
arXiv:2609.06573v1 Announce Type: cross Abstract: Recent studies report that automated red-teaming finds more vulnerabilities, at lower cost, than human red-teaming on standard AI safety benchmarks, and some read this as evidence that human evaluators are becoming dispensable. The comparison measures one thing and the conclusion claims another. A benchmark measures how thoroughly an attacker searches a predefined set of harms, fixed in advance by the developers, and a harm left out of that set is invisible to any attacker working inside it, automated or not. The same blind spot appeared in academic cryptography and in clinical drug trials, where an evaluation that was internally valid stayed silent about the population it was never pointed at. We call the AI-safety version the \emph{threat-model coverage gap}, and find that it persists in a current open-weight model, where harms surface in non-English prompts that English benchmarks miss. Closing it requires evaluators whose deployment context differs from the developers'. The case for those evaluators is methodological, grounded in coverage, and the existing evaluation frame is unlikely to produce them on its own.
Abstract:Recent studies report that automated red-teaming finds more vulnerabilities, at lower cost, than human red-teaming on standard AI safety benchmarks, and some read this as evidence that human evaluators are becoming dispensable. The comparison measures one thing and the conclusion claims another. A benchmark measures how thoroughly an attacker searches a predefined set of harms, fixed in advance by the developers, and a harm left out of that set is invisible to any attacker working inside it, automated or not. The same blind spot appeared in academic cryptography and in clinical drug trials, where an evaluation that was internally valid stayed silent about the population it was never pointed at. We call the AI-safety version the \emph{threat-model coverage gap}, and find that it persists in a current open-weight model, where harms surface in non-English prompts that English benchmarks miss. Closing it requires evaluators whose deployment context differs from the developers'. The case for those evaluators is methodological, grounded in coverage, and the existing evaluation frame is unlikely to produce them on its own.
Accepted for the 2nd Workshop on Safe AI at UAI (SafeAI 2026). 6 pages
Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Cryptography and Security (cs.CR); Computers and Society (cs.CY)
arXiv:2609.06573 [cs.AI]
(or arXiv:2609.06573v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2609.06573 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
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.
Text integral sub licența deschisă a publicației · 293 cuvinte
Subiecte conexe
[AINews] Reality Checks on AI News (Yegge shuts down Gas Town, Databricks’ +60% Astra cost)
Latent Space · 17 sept. 2026
State of the Field: AI for Epistemics and Coordination
LessWrong · 2 oct. 2026
A Red Line and Oversight Framework for Government AI Contracts
AI Alignment Forum · 18 iul. 2026
Latent reasoning architectures would likely undermine CoT, our strongest oversight tool
AI Alignment Forum · 23 sept. 2026