mlsanityData to 14 Oct 2026

Adversarial Prompts for Acceptance Collapse in Speculative Decoding

arXiv 2607.21804 ↗First posted 23 Jul 2026cs.CR cs.CL cs.LGcited by 1

On the map: Efficient LLM inference and serving → Speculative decoding

also in 1 more
Abstract (arXiv)

Lossless acceleration schemes, such as speculative decoding, promise significant inference speedups by relying on dynamic token-level alignment between a draft and a target model. However, this guarantee of semantic equivalence masks a severe operational vulnerability: draft-target alignment can be systematically attacked. In this paper, we introduce ADSD, which, to the best of our knowledge, is the first prompt-suffix attack that collapses verifier acceptance by pushing draft probability mass toward tokens the target is unlikely to accept. ADSD uses Soft-Collapse, a verifier-aligned surrogate derived from the asymmetric speculative acceptance rule, together with a target-preservation objective that discourages obvious task corruption. ADSD successfully generates highly effective adversarial suffixes. On the GSM8K dataset, our attack increases the mean sample time by 62.3% while preserving the task quality. We further show that this vulnerability exists across different domains, speculative decoding strategies, and model architectures.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Speculative decoding, one of 12 directions in Efficient LLM inference and serving. Builds on 46 of its 55 references that are in the map. First posted on arXiv 23 Jul 2026.

Builds on

46 of 55 references.

Closest ancestors · 23

Same direction, newest first

  1. 2026Speculative Speculative Decoding3 Mar
  2. 2026Overcoming Joint Intractability with Lossless Hierarchical Speculative Decoding9 Jan
  3. 2025Speculative Decoding: Performance or Illusion?31 Dec
  4. 2025Speculative Safety-Aware Decoding25 Aug
  5. 2025Alignment-Augmented Speculative Decoding with Alignment Sampling and Conditional Verification19 May
  6. 2025AdaSpec: Adaptive Speculative Decoding for Fast, SLO-Aware Large Language Model Serving7 Mar
  7. 2025EAGLE-3: Scaling up Inference Acceleration of Large Language Models via Training-Time Test3 Mar
  8. 2025Judge Decoding: Faster Speculative Sampling Requires Going Beyond Model Alignment31 Jan
  9. 2024When Speculation Spills Secrets: Side Channels via Speculative Decoding In LLMs1 Nov
  10. 2024AdaEDL: Early Draft Stopping for Speculative Decoding of Large Language Models via an Entropy-based Lower Bound on Token Acceptance Probability24 Oct
  11. 2024Optimized Multi-Token Joint Decoding with Auxiliary Model for LLM Inference12 Jul
  12. 2024EAGLE-2: Faster Inference of Language Models with Dynamic Draft Trees24 Jun
11 more
  1. 2024SDSAT: Accelerating LLM Inference through Speculative Decoding with Semantic Adaptive Tokens27 Mar
  2. 2024Block Verification Accelerates Speculative Decoding15 Mar
  3. 2024Direct Alignment of Draft Model for Speculative Decoding with Chat-Fine-Tuned LLMs29 Feb
  4. 2024Decoding Speculative Decoding2 Feb
  5. 2024EAGLE: Speculative Sampling Requires Rethinking Feature Uncertainty26 Jan
  6. 2024Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads19 Jan
  7. 2024Unlocking Efficiency in Large Language Model Inference: A Comprehensive Survey of Speculative Decoding15 Jan
  8. 2023Online Speculative Decoding11 Oct
  9. 2023Accelerating Large Language Model Decoding with Speculative Sampling2 Feb
  10. 2022Fast Inference from Transformers via Speculative Decoding30 Nov
  11. 2022Speculative Decoding: Exploiting Speculative Execution for Accelerating Seq2seq Generation30 Mar

Other references · 23

  1. 2023GPT-4 Technical Reportcited by 18,396
  2. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 13,028
  3. 2021Training Verifiers to Solve Math Word Problemscited by 10,109
  4. 2021Evaluating Large Language Models Trained on Codecited by 8,320
  5. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  6. 2023Qwen Technical Reportcited by 3,344
  7. 2023Universal and Transferable Adversarial Attacks on Aligned Language Modelscited by 3,104
  8. 2018The Lottery Ticket Hypothesis: Finding Sparse, Trainable Neural Networkscited by 2,262
  9. 2022SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Modelscited by 1,386
  10. 2017Get To The Point: Summarization with Pointer-Generator Networkscited by 1,292
  11. 2023A Simple and Effective Pruning Approach for Large Language Modelscited by 800
  12. 2023Baseline Defenses for Adversarial Attacks Against Aligned Language Modelscited by 651
11 more
  1. 2023Detecting Language Model Attacks with Perplexitycited by 431
  2. 2019Q-BERT: Hessian Based Ultra Low Precision Quantization of BERTcited by 359
  3. 2024SafeDecoding: Defending against Jailbreak Attacks via Safety-Aware Decodingcited by 247
  4. 2023LLM Self Defense: By Self Examination, LLMs Know They Are Being Trickedcited by 244
  5. 2020Sponge Examples: Energy-Latency Attacks on Neural Networkscited by 111
  6. 2023The Art of Defending: A Systematic Evaluation and Analysis of LLM Defense Strategies on Safety and Over-Defensivenesscited by 67
  7. 2024An Engorgio Prompt Makes Large Language Model Babble oncited by 30
  8. 2024LLM Safeguard is a Double-Edged Sword: Exploiting False Positives for Denial-of-Service Attackscited by 6
  9. 2024Amphista: Bi-directional Multi-head Decoding for Accelerating LLM Inferencecited by 5
  10. 2025LiteVLM: A Low-Latency Vision-Language Model Inference Pipeline for Resource-Constrained Environmentscited by 4
  11. 2026Rethinking Latency Denial-of-Service: Attacking the LLM Serving Framework, Not the Modelcited by 2

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Speedbumps: Rejection Attacks on Speculative Decodingcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Mistletoe: Stealthy Acceleration-Collapse Attacks on Speculative Decoding
  2. 2025Speculative Verification: Exploiting Information Gain to Refine Speculative Decoding
  3. 2026Evaluating Losslessness in Speculative Decoding Under Finite-Precision Inference
  4. 2026LK Losses: Direct Acceptance Rate Optimization for Speculative Decoding
  5. 2026Approximate Speculative Decoding
  6. 2026From Positionwise Confidence to Prefix Scheduling: Verifier Skipping in Speculative Decoding
  7. 2026PARD-2: Target-Aligned Parallel Draft Model for Dual-Mode Speculative Decoding
  8. 2026Acceptance-Aware Draft Model Training for Speculative Decoding