mlsanityData to 14 Oct 2026

When Can Conformal Risk Control Certify LLM Outputs? Bounds, Impossibility, and Adaptation for Structured Generation

arXiv 2606.29054 ↗First posted 27 Jun 2026cs.LGcited by 4

Not on the map. Why?

Abstract (arXiv)

Large language models (LLMs) deployed for structured generation (NER, JSON extraction, QA, and classification) lack formal reliability guarantees, and standard heuristic abstention policies violate user-specified risk targets on 7.5--12.5% of settings, with no per-deployment guarantee. We characterize when conformal risk control (CRC) can certify structured LLM outputs and when it provably cannot. First, we prove a sharpened, attained feasibility frontier: when base risk μ exceeds target α, any distribution-free method must abstain on at least (μ-α)/(M-α) of inputs, yielding a closed-form feasibility test that decides whether CRC can work before running it. Second, we establish a proven certification phase diagram across Hoeffding, empirical Bernstein, and a betting-based e-CRC bound, verified over 716 configurations (six open-weight models 3B--72B, eight datasets, six scores): at strict targets (α <= 0.20) the certified sets are nested (51/72/80 certified; Hoeffding-to-Bernstein the largest upgrade, +41%; e-CRC best under calibration scarcity), while at relaxed targets the Hoeffding-Bernstein ordering reverses at a closed-form frontier. Third, we show a negative shift result with a constructive counterpart: under cross-dataset shift the target is violated on 14 of 16 transfers by static CRC and every tested adaptive-conformal-inference (ACI) step size, yet a full-feedback anytime-valid monitor certifies 0 of 16 while remaining non-vacuous. Relaxing the target to α = 0.40 unlocks practical certification (28% NER, 13% QA, 19% CLS). The framework gives a three-step deployment recipe: check feasibility, select the bound and score, then re-check under shift.

Key facts. Cited by 4 papers in the map (data through 14 Oct 2026). Builds on 41 of its 56 references that are in the map. First posted on arXiv 27 Jun 2026.

Builds on

41 of 56 references.

References · 41

  1. 2020Measuring Massive Multitask Language Understandingcited by 7,963
  2. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  3. 2022Self-Consistency Improves Chain of Thought Reasoning in Language Modelscited by 5,708
  4. 2017On Calibration of Modern Neural Networkscited by 4,960
  5. 2024Qwen2.5 Technical Reportcited by 4,942
  6. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,955
  7. 2022Language Models (Mostly) Know What They Knowcited by 1,856
  8. 2023Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMscited by 974
  9. 2017Selective Classification for Deep Neural Networkscited by 852
  10. 2023Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generationcited by 819
  11. 2019Conformal Prediction Under Covariate Shiftcited by 519
  12. 2021Adaptive Conformal Inference Under Distribution Shiftcited by 375
29 more
  1. 2022Conformal prediction beyond exchangeabilitycited by 355
  2. 2019SelectiveNet: A Deep Neural Network with an Integrated Reject Optioncited by 350
  3. 2022Conformal Risk Controlcited by 299
  4. 2021Distribution-Free, Risk-Controlling Prediction Setscited by 288
  5. 2021Learn then Test: Calibrating Predictive Algorithms to Achieve Risk Controlcited by 271
  6. 2018Time-uniform, nonparametric, nonasymptotic confidence sequencescited by 229
  7. 2023Conformal Language Modelingcited by 170
  8. 2020Estimating means of bounded random variables by bettingcited by 169
  9. 2022Game-theoretic statistics and safe anytime-valid inferencecited by 155
  10. 2021Few-NERD: A Few-Shot Named Entity Recognition Datasetcited by 152
  11. 2024Language Models with Conformal Factuality Guaranteescited by 131
  12. 2024Trust or Escalate: LLM Judges with Provable Guarantees for Human Agreementcited by 105
  13. 2019Safe Testingcited by 101
  14. 2024Large language model validity via enhanced conformal prediction methodscited by 92
  15. 2021Optimal strategies for reject option classifierscited by 61
  16. 2024Mitigating LLM Hallucinations via Conformal Abstentioncited by 59
  17. 2024Conformal Alignment: Knowing When to Trust Foundation Models with Guaranteescited by 54
  18. 2024Conformal Prediction for Natural Language Processing: A Surveycited by 53
  19. 2023Non-Exchangeable Conformal Risk Controlcited by 29
  20. 2023Selective Generation for Controllable Language Modelscited by 15
  21. 2024Active, anytime-valid risk controlling prediction setscited by 14
  22. 2025Conformal Arbitrage: Risk-Controlled Balancing of Competing Objectives in Language Modelscited by 12
  23. 2025Cascaded Language Models for Cost-effective Human-AI Decision-Makingcited by 11
  24. 2025Selective Conformal Risk Controlcited by 11
  25. 2026Conformal Selective Acting: Anytime-Valid Risk Control for RLVR-Trained LLMscited by 8
  26. 2024Aligning Model Properties via Conformal Risk Controlcited by 7
  27. 2026PASC: Pipeline-Aware Conformal Prediction with Joint Coverage Guarantees for Multi-Stage NLP and LLM Pipelinescited by 7
  28. 2026Anytime-Valid Conformal Risk Controlcited by 5
  29. 2025Classification with Reject Option: Distribution-free Error Guarantees via Conformal Predictioncited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 4.

4 papers.

  1. 2026Non-Degenerate Risk Certification for Automated Security Decisions: A Decision-Contract Theory with ATT&CK-Aligned Triage as a Worked Instancecited by 0
  2. 2026The Exceedance Design Effect: Effective Sample Size for Thresholds under Clusteringcited by 0
  3. 2026Uncertainty Quantification for LLM Agents: A Taxonomy, an Evaluation Protocol, and an Empirical Studycited by 0
  4. 2026Certified Against Which Oracle? Execution Labels Set the Reported Risk of Conformal Abstention for Text-to-SQLcited by 0