The results indicate that static, benign-traffic-calibrated thresholds are insufficient for this defense, and that jitter-forgiveness thresholds should instead be calibrated dynamically against local token entropy.
Abstract
Gumbel-based inference verification bounds LLM weight exfiltration by only forgiving token choices that plausibly arise from honest GPU nondeterminism, reporting a>200x slowdown for a steganographic adversary under benign prompt traffic. This bound assumes a passive attacker; we show it degrades sharply against an adversary who instead controls the prompt distribution. Because the verifier's admissible-token-set size is driven by the model's own output entropy, prompts engineered to break grammatical and sub-word structure -- rather than benign conversational traffic -- widen that set and open a materially larger covert channel. Across six instruction-tuned models spanning 1B to 32B parameters and three random seeds, our strongest attack (character- and script-level disruption) roughly doubles bits leaked per token relative to benign prompts, cutting the slowdown factor to 60x - 118x. These results indicate that static, benign-traffic-calibrated thresholds are insufficient for this defense, and that jitter-forgiveness thresholds should instead be calibrated dynamically against local token entropy.
ADSD is introduced, which is the first prompt-suffix attack that collapses verifier acceptance by pushing draft probability mass toward tokens the target is unlikely to accept, and successfully generates highly effective adversarial suffixes.
Run-Min Wang, Chaoyi Zhou, Xi Liu et al.· arXiv.org· 0 citations
Deployed vision-language systems often gate their answers on confidence, making confidence robustness relevant to oversight. We study confidence readouts under white-box, image-only attacks constrained to preserve the generated answer byte-identically. Under a reachability assumption, an unmovable readout cannot outperform the answer-string accuracy prior, whose pooled value is 0.617. Independently of that assumption, a uniform amplitude certificate below a measurable threshold guarantees adversarial discrimination above the same floor. Across four vision-language models, three visual question answering benchmarks, five deployed confidence channels and two defense estimators, direct or surrogate-aimed attacks produce itemwise feasible perturbations that refute this uniform certificate in all 84 estimator-by-cell combinations. Coordinated correctness-label-aware attacks drive adversarial discrimination to or below the answer-string floor in all sixty deployed-channel cells, including all fifty-nine that begin above it. Hidden-state interventions and an open-ended text-model activation-space replication show that comparable confidence movement can be induced at the representation level rather than only through adversarial images. None of four tested defense families establishes a robust alternative under the specific evaluation applied to it. In a confidence-gated simulation, a coordinated token-probability attack transferred to a hidden-state gate causes up to 84.8% of previously rejected wrong answers to become accepted. After reweighting to each benchmark's natural correctness prevalence, accepted accuracy falls below the no-gate baseline in eight of twelve cells under transfer and all twelve under a direct gate-aimed attack. Under the studied threat model and budget, confidence is therefore an integrity-sensitive rather than intrinsically robust oversight signal.
Reza Khanmohammadi, Ivan Brugere, Simerjot Kaur et al.· 0 citations
A privacy-preserving zk-SNARK-based audit framework that searches for probes designed in the spirit of adversarial examples to amplify logit drift between an approved model and a modified deployment and demonstrates that token-based probes consistently deliver the strongest mean sensitivity across models and GPU platforms, although operating in a black-box setting.
Cameron Wilding, Mina Shaker, Fatemeh Ganji· 0 citations
We introduce Stateless Bernoulli Watermarking (SBW), a new statistical watermark for Large Language Models that determines green list membership through independent per-token Bernoulli trials. Unlike KGW's vocabulary permutation or SynthID's multi-layer tournament, SBW requires only a single comparison per token against a counter-based random number generator, reducing membership complexity to $O(1)$ and enabling single-kernel execution with zero intermediate allocations. We prove that this formulation preserves the same detection guarantees as fixed-size green lists: the z-score test remains $\mathcal{N}(0,1)$ under the null. The stateless architecture enables capabilities unavailable to existing methods: full-vocabulary self-salt watermarking (over 6000$\times$ faster than KGW's self-salt and 2$\times$ faster than SynthID despite biasing the entire vocabulary with candidate-dependent seeding) and architectural compatibility with distributed inference. In end-to-end generation benchmarks, SBW adds less than 1\% overhead at all batch sizes. We additionally identify hash function design as a previously unexplored axis for watermark quality, showing that a GPU-native Jenkins hash improves null calibration by 1.8$\times$ while producing more diverse text. Experiments across two seeding schemes and eight $(\gamma, \delta)$ configurations confirm statistical equivalence with ROC-AUC differences below 0.01.
HEbdMIA is a lightweight homomorphic encryption-based defense that operates at the post-inference stage by encrypting model output logits without requiring retraining or architectural modifications, indicating that HEbdMIA provides an effective, scalable, and deployment-friendly solution for enhancing privacy in real-world machine learning systems.
Akash Shah, M. A. Wani, R. Chaturvedi et al.· Computers, Materials & C...· 0 citations
TriShield is presented, a three-layer deterministic defense that completely prevents NeuroImprint-style reconstruction with zero model utility loss and no additional communication rounds, and it is proved theoretically that after Layers 2 and 3, the mutual information between the uploaded gradient and any individual training sample is zero.
Chenghui Wei· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
