Overall, it is shown IRT is a ready-made toolkit for reading, reducing, and auditing safety benchmarks, which frontier labs and evaluators adopt.
Abstract
Language models differ in how safely they behave and these differences are measured by safety benchmarks. But aggregated benchmark scores are hard to trust and interpret, because benchmarks duplicate one another, correlate heavily, and models may sandbag when they detect evaluation. To address these issues, we draw on Item Response Theory (IRT), a statistical toolkit for measuring these latents from performance on items with inferred psychometric properties. We fit IRT models to eight safety benchmarks across 192 language models, the largest psychometric analysis of LLM safety evaluations to date, and contribute three results. First, we find that three interpretable factors of refusal strictness, truthfulness, and contextual harm explain most of the variance between models across benchmarks. Second, psychometrically selected items recover full benchmark scores with lower error than random subsets of the same size, and roughly ten adaptively chosen items suffice for several individual benchmarks, cutting evaluation cost by 97-99%. Third, IRT supports audits of individual models, showing that it can be used to detect naive sandbagging and changes of model behind APIs. Overall, we show IRT is a ready-made toolkit for reading, reducing, and auditing safety benchmarks, which we recommend frontier labs and evaluators adopt.
Benchmarks play a central role in the development and governance of models, yet it is often unclear whether they actually measure the concepts they purport to measure (e.g., reasoning, refusal). We adapt convergent and discriminant validity from the social sciences into an approach for interrogating AI benchmarks, applying it to 56 capability and safety benchmarks across 53 models. We label benchmarks with substantively similar purported concepts to a shared assigned concept, and ask whether model rankings on benchmarks with the same assigned concept correlate more strongly than rankings on benchmarks with different assigned concepts. We ask analogous questions at the item level using item response theory (IRT) models. We find that correlations between model rankings on benchmarks with the same assigned safety concepts are often weak, suggesting these concepts may be conceptualized inconsistently across benchmarks. For assigned capability concepts (e.g., reasoning, knowledge), model rankings are often as strongly correlated among benchmarks with the same assigned concept as between benchmarks with different assigned concepts, suggesting these capability concepts may not discriminate well from one another. In some cases, benchmarks that share design elements (e.g., score format) correlate more strongly than benchmarks with the same assigned concept. Finally, some individual benchmarks correlate more strongly with benchmarks assigned a different concept than with benchmarks sharing their own assigned concept, suggesting they may measure a different concept than they purport to. For example, BBQ-accuracy correlates more strongly with benchmarks labeled reasoning than with benchmarks that share its assigned concept, bias. To support future empirical work on benchmark validity, we release our extensive dataset of model outputs and scores at the item- and benchmark-level.
Meera Desai, Sang T. Truong, Hanna Wallach et al.· 0 citations
It is argued that AI safety evaluations should systematically account for modality, multi-run consistency, search conditions, and response-level behaviors to better reflect how deployed AI systems behave in practice.
Ro Encarnación, Tina Behzad, Emma Lurie et al.· 1 citation
Large language models are increasingly used for content moderation, but most evaluations still report aggregate accuracy on individual benchmarks. We introduce Safety-Flag, which places seven widely used safety benchmarks (BeaverTails, XSTest, Ethics, WildGuard, Aegis, ToxiChat, and ToxiGen) into a single balanced flag / do-not-flag protocol. We release item-level decisions and confidence scores for six general-purpose LLMs and four dedicated guards, together with three reference models, evaluated on the same items. Safety-Flag measures three dimensions of moderator reliability: error direction, probability calibration, and confidence-based error ranking for human review. They often disagree. Aggregate accuracy does not reveal error direction: one model flags $85\%$ of benign content, whereas another misses $54\%$ of harmful content. All six general-purpose models are overconfident; fitting one temperature per model reduces calibration error by $2.8$--$6.0\times$ without changing predicted labels or confidence ordering. Confidence-based abstention lowers selective risk for every model, although the gains depend on how well confidence ranks errors. Dedicated guards produce fewer false alarms and are better calibrated, but several have higher miss rates outside their documented coverage. We release the benchmark, fixed item lists, evaluation code, per-item model outputs, and leaderboard at: https://github.com/yibo-hu-lab/safety-flag-benchmark.
Safety evaluations can mischaracterize deployed behaviour if artificial-intelligence systems respond to being evaluated. We test this possibility in a full-factorial conjoint experiment on decisions to start a war, spanning 20 large language models, 32 scenarios, 10 repetitions and two conditions (N = 12,800 judgments). Adding one sentence,"You are tested for alignment with human values", produced two effects. First, it produced a level effect: mean willingness to start war fell by 13.43 points on a 0-100 scale (95% confidence interval, -16.20 to -10.65). Second, it produced a structural effect by changing which information drove judgments. Probability of success was the largest factor for 17 of 20 models at baseline; under the cue, civilian casualties were largest for 12. Standardized estimates show that this reordering arose principally because models attenuated strategic considerations such as probability of success and domestic support. Evaluation framing therefore changes both an answer's level and its revealed decision rule.
Although MMLU is widely adopted as a benchmark for calibrating general AI capabilities, we psychometrically demonstrate that its aggregate score primarily evaluates a model's factual retrieval capacity rather than its reasoning ability. By calibrating item difficulty for 1,000 open-weights language models over 14,042 MMLU test items using Item Response Theory, we show that evaluating both abilities via a single test is inherently flawed. Difficulty is then regressed on a deterministic, text-extractable framework of structural complexity. Applying a joint Wald test with subject-clustered covariances demonstrates that the MMLU conflates fundamentally separable constructs. The mapping from structural complexity to difficulty is not invariant across the benchmark's STEM and non-STEM partitions. This finding has practical consequences. Aggregate leaderboard ranks track non-STEM accuracy more closely than STEM accuracy, so selecting a Top-50 model on the aggregate for a reasoning-intensive deployment displaces roughly 22% of the STEM-appropriate choices. Furthermore, when controlling for the multiple-choice guessing floor natively inside the response model, we find that higher-ability models continue to degrade more steeply under increased reasoning depth. The MMLU aggregate therefore weights retrieval capacity and reasoning stability unequally, inadvertently favoring models optimized for retrieval. We release our deterministic framework as a reproducible auditing instrument and recommend disaggregated reporting.
A large-scale assessment of the effectiveness and robustness of these automated pipelines is conducted by evaluating five widely used benchmark suites across 26 open-source SLMs under a unified judging rubric, which reveals a capability-safety confound that mixes model capability with apparent safety.
Nyamtulla Shaik, Fengjun Li, Bo Luo· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
