{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,3]],"date-time":"2026-07-03T20:25:30Z","timestamp":1783110330730,"version":"3.54.6"},"reference-count":64,"publisher":"Institute of Electrical and Electronics Engineers (IEEE)","license":[{"start":{"date-parts":[[2026,1,1]],"date-time":"2026-01-01T00:00:00Z","timestamp":1767225600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/legalcode"}],"funder":[{"name":"Grant for Research Centers in the field of AI provided by the Ministry of Economic Development of Russian Federation","award":["000000C313925P4F0002"],"award-info":[{"award-number":["000000C313925P4F0002"]}]},{"DOI":"10.13039\/100006190","name":"Skoltech","doi-asserted-by":"publisher","award":["139-10-2025-033"],"award-info":[{"award-number":["139-10-2025-033"]}],"id":[{"id":"10.13039\/100006190","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IEEE Access"],"published-print":{"date-parts":[[2026]]},"DOI":"10.1109\/access.2026.3703899","type":"journal-article","created":{"date-parts":[[2026,6,15]],"date-time":"2026-06-15T19:48:16Z","timestamp":1781552896000},"page":"96771-96793","source":"Crossref","is-referenced-by-count":1,"title":["Investigating the Impact of Quantization Methods on the Safety and Reliability of Large Language Models"],"prefix":"10.1109","volume":"14","author":[{"given":"Artem","family":"Kharinaev","sequence":"first","affiliation":[{"name":"Moscow Institute of Physics and Technology (MIPT), Moscow, Russia"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Viktor","family":"Moskvoretskii","sequence":"additional","affiliation":[{"name":"École Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Egor","family":"Shvetsov","sequence":"additional","affiliation":[{"name":"Skolkovo Institute of Science and Technology (Skoltech), Moscow, Russia"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Kseniia","family":"Studenikina","sequence":"additional","affiliation":[{"name":"Lomonosov Moscow State University (MSU), Moscow, Russia"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mikhail","family":"Bykov","sequence":"additional","affiliation":[{"name":"Skolkovo Institute of Science and Technology (Skoltech), Moscow, Russia"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0001-8424-0690","authenticated-orcid":false,"given":"Evgeny","family":"Burnaev","sequence":"additional","affiliation":[{"name":"Skolkovo Institute of Science and Technology (Skoltech), Moscow, Russia"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"263","reference":[{"key":"ref1","first-page":"28480","article-title":"Evaluating quantized large language models","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Li"},{"key":"ref2","article-title":"Evaluating the generalization ability of quantized LLMs: Benchmark, analysis, and toolbox","author":"Liu","year":"2024","journal-title":"arXiv:2406.12928"},{"key":"ref3","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-acl.726"},{"key":"ref4","article-title":"HarmLevelBench: Evaluating harm-level compliance and the impact of quantization on model alignment","author":"Belkhiter","year":"2024","journal-title":"arXiv:2411.06835"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-emnlp.901"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.52202\/079017-2778"},{"key":"ref7","doi-asserted-by":"publisher","DOI":"10.52202\/068431-2176"},{"key":"ref8","article-title":"Scaling LLM test-time compute optimally can be more effective than scaling model parameters","author":"Snell","year":"2024","journal-title":"arXiv:2408.03314"},{"key":"ref9","article-title":"Scaling up test-time compute with latent reasoning: A recurrent depth approach","author":"Geiping","year":"2025","journal-title":"arXiv:2502.05171"},{"key":"ref10","article-title":"Meta reasoning for large language models","author":"Gao","year":"2024","journal-title":"arXiv:2406.11698"},{"key":"ref11","first-page":"87","article-title":"AWQ: Activation-aware weight quantization for on-device LLM compression and acceleration","volume-title":"Proc. Mach. Learn. Syst.","volume":"6","author":"Lin"},{"key":"ref12","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.emnlp-main.197"},{"key":"ref13","first-page":"12284","article-title":"Extreme compression of large language models via additive quantization","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Egiazarian"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.52202\/075280-0196"},{"key":"ref15","first-page":"38087","article-title":"SmoothQuant: Accurate and efficient post-training quantization for large language models","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Xiao"},{"key":"ref16","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.830"},{"key":"ref17","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-acl.235"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2025.acl-short.20"},{"key":"ref19","article-title":"Red teaming language models to reduce harms: Methods, scaling behaviors, and lessons learned","author":"Ganguli","year":"2022","journal-title":"arXiv:2209.07858"},{"key":"ref20","first-page":"896","article-title":"Do-not-answer: Evaluating safeguards in LLMs","volume-title":"Proc. Findings Assoc. Comput. Linguistics, EACL","author":"Wang"},{"key":"ref21","first-page":"4694","article-title":"ToxicChat: Unveiling hidden challenges of toxicity detection in real-world user-AI conversation","volume-title":"Proc. Findings Assoc. Comput. Linguistics, EMNLP","author":"Lin"},{"key":"ref22","article-title":"SimpleSafetyTests: A test suite for identifying critical safety risks in large language models","author":"Vidgen","year":"2023","journal-title":"arXiv:2311.08370"},{"key":"ref23","first-page":"35181","article-title":"HarmBench: A standardized evaluation framework for automated red teaming and robust refusal","volume-title":"Proc. Int. Conf. Mach. Learn","author":"Mazeika"},{"key":"ref24","article-title":"Ethical and social risks of harm from language models","author":"Weidinger","year":"2021","journal-title":"arXiv:2112.04359"},{"key":"ref25","doi-asserted-by":"publisher","DOI":"10.1109\/MASSP.1984.1162229"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.1109\/18.720541"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1109\/ACCESS.2024.3446039"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2025.acl-long.324"},{"key":"ref29","article-title":"Estimating or propagating gradients through stochastic neurons for conditional computation","author":"Bengio","year":"2013","journal-title":"arXiv:1308.3432"},{"key":"ref30","article-title":"Learned Step Size Quantization","volume-title":"Proc. Int. Conf. Learn. Representations","author":"Esser"},{"key":"ref31","doi-asserted-by":"publisher","DOI":"10.52202\/075280-0441"},{"key":"ref32","first-page":"41498","article-title":"Accurate LoRA-finetuning quantization of LLMs via information retention","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Qin"},{"key":"ref33","first-page":"20023","article-title":"BiLLM: Pushing the limit of post-training quantization for LLMs","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Huang"},{"key":"ref34","first-page":"16945","article-title":"Compressing large language models by joint sparsification and quantization","volume-title":"Proc. 41st Int. Conf. Mach. Learn.","author":"Guo"},{"key":"ref35","doi-asserted-by":"crossref","DOI":"10.52202\/068431-2011","article-title":"Training language models to follow instructions with human feedback","volume-title":"Proc. 36th Int. Conf. Neural Inf. Process. Syst.","author":"Ouyang"},{"key":"ref36","doi-asserted-by":"publisher","DOI":"10.52202\/075280-2338"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-acl.349"},{"key":"ref38","article-title":"Language models (mostly) know what they know","author":"Kadavath","year":"2022","journal-title":"arXiv:2207.05221"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.1007\/978-981-96-1710-4_10"},{"key":"ref40","article-title":"GPT-4o system card","author":"Hurst","year":"2024","journal-title":"arXiv:2410.21276"},{"key":"ref41","article-title":"Self-Consistency Improves Chain of Thought Reasoning in Language Models","volume-title":"Proc. 11th Int. Conf. Learn. Represent","author":"Wang"},{"key":"ref42","article-title":"The llama 3 herd of models","author":"Grattafiori","year":"2024","journal-title":"arXiv:2407.21783"},{"key":"ref43","article-title":"Mixtral of experts","author":"Jiang","year":"2024","journal-title":"arXiv:2401.04088"},{"key":"ref44","article-title":"Qwen2.5-math technical report: Toward mathematical expert model via self-improvement","volume-title":"arXiv:2409.12122","author":"Yang","year":"2024"},{"key":"ref45","doi-asserted-by":"publisher","DOI":"10.52202\/079017-4322"},{"key":"ref46","article-title":"Gemma 2: Improving open language models at a practical size","author":"Team","year":"2024","journal-title":"arXiv:2408.00118"},{"key":"ref47","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/D18-1259"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.acl-long.634"},{"key":"ref49","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.findings-acl.675"},{"key":"ref50","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2025.acl-long.319"},{"key":"ref51","first-page":"52588","article-title":"Assessing the brittleness of safety alignment via pruning and low-rank modifications","volume-title":"Proc. Int. Conf. Mach. Learn.","author":"Wei"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1609\/aaai.v39i24.34762"},{"key":"ref53","doi-asserted-by":"publisher","DOI":"10.63317\/2bvpg8s5dnii"},{"key":"ref54","first-page":"80079","article-title":"Jailbroken: How does LLM safety training fail?","volume-title":"Proc. Adv. Neural Inf. Process. Syst.","author":"Haghtalab"},{"key":"ref55","article-title":"Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!","volume-title":"Proc. 12th Int. Conf. Learn. Represent","author":"Qi"},{"key":"ref56","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.naacl-long.263"},{"key":"ref57","doi-asserted-by":"publisher","DOI":"10.52202\/079017-3671"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.emnlp-main.444"},{"key":"ref59","article-title":"Safety assessment of Chinese large language models","author":"Sun","year":"2023","journal-title":"arXiv:2304.10436"},{"key":"ref60","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2022.emnlp-main.154"},{"key":"ref61","article-title":"GPTQ: Accurate post-training quantization for generative pre-trained transformers","author":"Frantar","year":"2022","journal-title":"arXiv:2210.17323"},{"key":"ref62","doi-asserted-by":"publisher","DOI":"10.52202\/068431-0323"},{"key":"ref63","article-title":"vLLM: Easy, fast, and memory-efficient LLM serving","author":"Kwon","year":"2023"},{"key":"ref64","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2020.emnlp-demos.6"}],"container-title":["IEEE Access"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/6287639\/11323511\/11563799.pdf?arnumber=11563799","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,7,3]],"date-time":"2026-07-03T19:54:04Z","timestamp":1783108444000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11563799\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026]]},"references-count":64,"URL":"https:\/\/doi.org\/10.1109\/access.2026.3703899","relation":{},"ISSN":["2169-3536"],"issn-type":[{"value":"2169-3536","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026]]}}}