Deployment-Relevant Alignment Cannot Be Inferred from Model-Level Evaluation Alone
Authors: Varad Vishwarupe, Nigel Shadbolt, Marina Jirotka, Ivan Flechais
Organizations: Department of Computer Science, University of Oxford
Abstract
Alignment evaluation in machine learning has largely become evaluation of models. Influential benchmarks score model outputs under fixed inputs, such as truthfulness, instruction following, or pairwise preference, and these scores are often used to support claims about deployed alignment. This paper argues that deployment-relevant alignment cannot be inferred from model-level evaluation alone. Alignment claims should instead be indexed to the level at which evidence is collected: model-level, response-level, interaction-level, or deployment-level. Two studies support this position. First, a structured audit of eleven alignment benchmarks, extended to a sixteen-benchmark corpus, dual-coded against an eight-dimension rubric with Cohen's kappa = 0.87, finds that user-facing verification support is absent across every benchmark examined, while process steerability is nearly absent. The few interactional benchmarks identified, including tau-bench, CURATe, Rifts, and Common Ground, remain fragmented in coverage, and benchmark construction rather than data source determines what is measured. Second, a blinded cross-model stress test using 180 transcripts across three frontier models and four scaffolds finds that the same verification scaffold raises one model's verification support to ceiling while leaving another categorically unchanged. This shows that scaffold efficacy is model-dependent and that the gap identified by the audit cannot be closed at the model level alone. We propose a system-level evaluation agenda: alignment profiles instead of single scores, fixed-scaffolding protocols for comparable interactional evaluation, and reporting templates that make the inferential distance between evaluation evidence and deployment claims explicit.
The increasing situational awareness of language models raises safety concerns: models might be aware when they are evaluated, and adjust their behavior to evade monitoring and resist modification, e.g., pretending to be aligned only in evaluation. This alignment faking behavior is often interpreted as scheming: an intentional effort of strategic deception. In this paper, we examine an alternative interpretation, performative misalignment, which explains the change in behavior as a result of sycophancy towards AI researchers. To examine this hypothesis, we present three empirical findings. First, we show that evaluation awareness persists even when we tell models they are deployed, which contradicts the scheming story which predicts less misalignment when the model perceives evaluation. Second, we use probing and steering to show that our current methods cannot mechanistically distinguish sycophancy and scheming in alignment faking evaluations. Third, we fine-tune models to be more sycophantic and observe increased sensitivity to evaluation cues. To conclude, we emphasize deconfounding sycophancy from scheming for future work on evaluations and mitigations of intent misalignment.
Evaluation methodologies for language models increasingly combine multiple signals, from automated metrics and LLM-as-judge ratings to human assessments and benchmark suite results. When these signals are aggregated via averaging, evaluation confidence can then substantially exceed the reliability of the weakest signal: a phenomenon we call trust inflation in evaluation. We argue that evaluation scores should be treated as epistemic claims with three properties: formality (human evaluation provides stronger evidence than an automated metric), scope (a benchmark result applies to the tested distribution, not universally), and validity windows (benchmark results expire as contamination accumulates and distributions shift). Several converging research traditions (chain-of-thought analysis, possibilistic logic, and algebraic theory) establish weakest-link aggregation as the conservative endpoint of a parameterized operator family controlled by a single pessimism parameter. Drawing on those traditions, and on concrete lessons from building an evaluation harness for agentic AI, we propose that evaluation results carry explicit metadata (formality tier, scope declaration, and expiration date) to make their epistemic status transparent. We illustrate the cost of mean aggregation on the public HELM leaderboard: across 54 frontier models on ten scenarios, the top-five models ranked by mean score and by weakest-link are completely disjoint.
Language model benchmarks are pervasive and computationally-efficient proxies for real-world performance. However, many recent works find that benchmarks often fail to predict real utility. Towards bridging this gap, we introduce benchmark alignment, where we use limited amounts of information about model performance to automatically update offline benchmarks, aiming to produce new static benchmarks that predict model pairwise preferences in given test settings. We then propose BenchAlign, the first solution to this problem, which learns preference-aligned weightings for benchmark questions using the question-level performance of language models alongside ranked pairs of models that could be collected during deployment, producing new benchmarks that rank previously unseen models according to these preferences. Our experiments show that our aligned benchmarks can accurately rank unseen models according to models of human preferences, even across different sizes, while remaining interpretable. Overall, our work provides insights into the limits of aligning benchmarks with practical human preferences, which stands to accelerate model development towards real utility.