Model stealing attacks, where adversaries create high-fidelity surrogate models, are a significant threat to the intellectual property of machine learning services. Conventional wisdom suggests these surrogates could provide adversaries with economic leverage comparable to the original service providers. This paper challenges this assumption by evaluating model stealing attacks beyond mere fidelity to the target model. Because query-based extraction provides only partial supervision of the target's input-output behavior, the surrogate is not uniquely identified: many near-optimal surrogates can achieve comparable fidelity while differing in deployment-relevant properties. Instead of performing a classic learning-based model stealing attack, we compute the Rashomon Set (i.e., the set of almost-equally-accurate models) of surrogate models, and evaluate its diversity using multiplicity metrics (ambiguity, discrepancy, and Rashomon Capacity) and group fairness metrics. Across tabular, medical imaging, and NLP tasks, our experiments on real-world datasets reveal that despite exhibiting similar fidelity to the target model, surrogate models can display significant variances in other critical performance metrics. These findings cast doubt on the presumed equivalence between high-fidelity surrogates and the target model in practical deployment scenarios.
We present Caliber, an output-perturbation defense against model extraction that formulates noise selection as a calibration problem: how much the defense degrades the supervision signal used to train a surrogate, and the provable per-input query cost of recovering the clean logits. To defend against an attacker that uses returned scores for knowledge distillation, Caliber adds independent and identically distributed Gaussian noise to the internal logits. We establish two properties of the resulting perturbed predictions. Monotone agreement degradation: When the clean logits have a unique maximizer, agreement with the clean prediction decreases strictly with the noise scale, so every target in (1/K,1) corresponds to a unique positive scale; task accuracy is bounded by computable lower and upper envelopes. Per-input recovery cost: We derive a closed-form minimax lower bound on the repeated queries needed to recover the clean logits for a fixed input. Caliber normalizes noise variance by the squared median top-two logit margin and fits the resulting noise-utility relationship with a logistic curve, either per model or shared within a task. Across more than thirty model-dataset combinations, per-model calibration achieves mean absolute relative errors of 0.6-1.4%. End-to-end experiments show that surrogate performance generally tracks the configured degradation, while fixed-input averaging follows the expected variance reduction.
Mechanistic interpretability (MI) requires full access to model internals, yet the APIs for most widely deployed language models at best expose log-probabilities over output tokens. This creates a surrogate problem: when do measurements made on open models allow us to make claims about a closed model? We evaluate surrogate fidelity at the prediction, attribution, and representation levels. For binary classification tasks, log-odds provide an API-compatible scalar readout of the model's representation space, and leave-one-out attributions provide insight into model behavior. Across eleven models spanning four families (Llama, Qwen, GPT, and Gemini), we find that prediction fidelity substantially overstates attribution fidelity: models that agree on what the answer is often disagree on why. We document an access-validity inversion: white-box signals like attention patterns and perturbation magnitudes are highly stable across models but only weakly predictive of causal attributions, which black-box input ablations capture by design. Mechanistic insight does not automatically transfer to closed targets, and prediction-level agreement is insufficient to warrant such transfer. Code and results are available at https://github.com/facebookresearch/surrogate.
Philippe Chlenski, Zachariah Carmichael, Ayush Warikoo +5
Counterfactual explanations (CFs) help understand machine learning models by identifying minimal input changes that would lead to alternative model outcomes. Recent work demonstrates their utility for reconstructing black-box models, enabling third-party auditing of opaque decision systems for fairness and accountability. Still, CF-based reconstruction may suffer from decision boundary shifts, overfitting, and restrictive assumptions requiring online query access to target platforms. We propose REconstruction via Counterfactual-Aware waSserstein opTimization (RECAST) under limited data and restricted access, a behavioral surrogate model based on Wasserstein barycentric prototypes. Our approach addresses decision boundary shifts by incorporating CFs as informative, though less representative, samples for both classes, maintaining high surrogate fidelity in low-sample regimes without requiring online access during reconstruction. To enhance fairness auditing, our method enables systematic group fairness diagnostics. Experiments on real-world datasets and various setups show that RECAST effectively achieves high fidelity and query efficiency, as well as stable results even when the access is limited and noisy.