mlsanityData to 14 Oct 2026

LLM Censorship: A Machine Learning Challenge or a Computer Security Problem?

arXiv 2307.10719 ↗First posted 20 Jul 2023cs.AI cs.CL cs.CRcited by 49

Not on the map. Why?

Cited by 21 in LLM safety and security, but placed by its own abstract, which didn’t fit.

Abstract (arXiv)

Large language models (LLMs) have exhibited impressive capabilities in comprehending complex instructions. However, their blind adherence to provided instructions has led to concerns regarding risks of malicious use. Existing defence mechanisms, such as model fine-tuning or output censorship using LLMs, have proven to be fallible, as LLMs can still generate problematic responses. Commonly employed censorship approaches treat the issue as a machine learning problem and rely on another LM to detect undesirable content in LLM outputs. In this paper, we present the theoretical limitations of such semantic censorship approaches. Specifically, we demonstrate that semantic censorship can be perceived as an undecidable problem, highlighting the inherent challenges in censorship that arise due to LLMs' programmatic and instruction-following capabilities. Furthermore, we argue that the challenges extend beyond semantic censorship, as knowledgeable attackers can reconstruct impermissible outputs from a collection of permissible ones. As a result, we propose that the problem of censorship needs to be reevaluated; it should be treated as a security problem which warrants the adaptation of security-based approaches to mitigate potential risks.

Key facts. Cited by 49 papers in the map (data through 14 Oct 2026). Builds on 29 of its 59 references that are in the map. First posted on arXiv 20 Jul 2023.

Builds on

29 of 59 references.

References · 29

  1. 2020Language Models are Few-Shot Learnerscited by 32,419
  2. 2022Training language models to follow instructions with human feedbackcited by 15,971
  3. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,182
  4. 2019Fine-Tuning Language Models from Human Preferencescited by 2,150
  5. 2021WebGPT: Browser-assisted question-answering with human feedbackcited by 1,745
  6. 2023Jailbroken: How Does LLM Safety Training Fail?cited by 1,651
  7. 2023Not what you've signed up for: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injectioncited by 1,443
  8. 2023PaLM 2 Technical Reportcited by 1,162
  9. 2022Red Teaming Language Models with Language Modelscited by 1,026
  10. 2022Discovering Language Model Behaviors with Model-Written Evaluationscited by 836
  11. 2022Ignore Previous Prompt: Attack Techniques For Language Modelscited by 832
  12. 2022Quantifying Memorization Across Neural Language Modelscited by 763
17 more
  1. 2023The Internal State of an LLM Knows When It's Lyingcited by 659
  2. 2023Eliciting Latent Predictions from Transformers with the Tuned Lenscited by 623
  3. 2023Jailbreaking ChatGPT via Prompt Engineering: An Empirical Studycited by 498
  4. 2023Augmented Language Models: a Surveycited by 403
  5. 2023Can AI-Generated Text be Reliably Detected?cited by 386
  6. 2022A Holistic Approach to Undesired Content Detection in the Real Worldcited by 356
  7. 2023Large Language Models as Tool Makerscited by 303
  8. 2023Exploiting Programmatic Behavior of LLMs: Dual-Use Through Standard Security Attackscited by 278
  9. 2023Fundamental Limitations of Alignment in Large Language Modelscited by 155
  10. 2023On the Tool Manipulation Capability of Open-source Large Language Modelscited by 122
  11. 2020Sponge Examples: Energy-Latency Attacks on Neural Networkscited by 111
  12. 2021Bad Characters: Imperceptible NLP Attackscited by 71
  13. 2022Evaluating the Susceptibility of Pre-Trained Language Models via Handcrafted Adversarial Examplescited by 69
  14. 2023Memory Augmented Large Language Models are Computationally Universalcited by 47
  15. 2022On the Impossible Safety of Large AI Modelscited by 28
  16. 2021Impossibility Results in AI: A Surveycited by 13
  17. 2023When Vision Fails: Text Attacks Against ViT and OCRcited by 5

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2023: 7; 2024: 25; 2025: 12; 2026: 5.

49 papers.

Sort:

By area:

  1. 2023Code Llama: Open Foundation Models for Codecited by 2,139
  2. 2024HarmBench: A Standardized Evaluation Framework for Automated Red Teaming and Robust Refusalcited by 1,429
  3. 2024StruQ: Defending Against Prompt Injection with Structured Queriescited by 375
  4. 2023Survey of Vulnerabilities in Large Language Models Revealed by Adversarial Attackscited by 194
  5. 2024LLM Defenses Are Not Robust to Multi-Turn Human Jailbreaks Yetcited by 151
  6. 2023Identifying and Mitigating the Security Risks of Generative AIcited by 91
  7. 2024Coercing LLMs to do and reveal (almost) anythingcited by 89
  8. 2023Use of LLMs for Illicit Purposes: Threats, Prevention Measures, and Vulnerabilitiescited by 82
  9. 2024Safeguarding Large Language Models: A Surveycited by 78
  10. 2024JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Modelscited by 70
  11. 2024Large Language Models for Cyber Security: A Systematic Literature Reviewcited by 68
  12. 2024Bag of Tricks: Benchmarking of Jailbreak Attacks on LLMscited by 51
37 more
  1. 2024JAILJUDGE: A Comprehensive Jailbreak Judge Benchmark with Multi-Agent Enhanced Explanation Evaluation Frameworkcited by 47
  2. 2023Can Language Models be Instructed to Protect Personal Information?cited by 46
  3. 2024Operationalizing a Threat Model for Red-Teaming Large Language Models (LLMs)cited by 46
  4. 2024Breaking Down the Defenses: A Comparative Survey of Attacks on Large Language Modelscited by 44
  5. 2024ASETF: A Novel Method for Jailbreak Attack on LLMs through Translate Suffix Embeddingscited by 32
  6. 2024Adversaries Can Misuse Combinations of Safe Modelscited by 32
  7. 2024MOSSBench: Is Your Multimodal Language Model Oversensitive to Safe Queries?cited by 30
  8. 2024Navigating LLM Ethics: Advancements, Challenges, and Future Directionscited by 28
  9. 2023Make Them Spill the Beans! Coercive Knowledge Extraction from (Production) LLMscited by 27
  10. 2024Khayyam Challenge (PersianMMLU): Is Your LLM Truly Wise to The Persian Language?cited by 22
  11. 2024Breach By A Thousand Leaks: Unsafe Information Leakage in `Safe' AI Responsescited by 18
  12. 2024Security Attacks on LLM-based Code Completion Toolscited by 16
  13. 2024Attack Atlas: A Practitioner's Perspective on Challenges and Pitfalls in Red Teaming GenAIcited by 16
  14. 2025Steering the CensorShip: Uncovering Representation Vectors for LLM "Thought" Controlcited by 14
  15. 2025What Large Language Models Do Not Talk About: An Empirical Study of Moderation and Censorship Practicescited by 11
  16. 2024Whose LLM is it Anyway? Linguistic Comparison and LLM Attribution for GPT-3.5, GPT-4 and Bardcited by 10
  17. 2023Moderating Model Marketplaces: Platform Governance Puzzles for AI Intermediariescited by 9
  18. 2024A Survey on Responsible Generative AI: What to Generate and What Notcited by 8
  19. 2024Exploring the Potential of Large Language Models for Improving Digital Forensic Investigation Efficiencycited by 7
  20. 2025Prompt, Divide, and Conquer: Bypassing Large Language Model Safety Filters via Segmented and Distributed Prompt Processingcited by 7
  21. 2024Conversational Complexity for Assessing Risk in Large Language Modelscited by 6
  22. 2024Generative AI Security: Challenges and Countermeasurescited by 4
  23. 2024Securing the Future of GenAI: Policy and Technologycited by 4
  24. 2026Emergence World: A Platform for Evaluating Long-Horizon Multi-Agent Autonomycited by 4
  25. 2026What Can Be Enforced? A Theory of Certified Runtime Safety for Tool-Using Agentscited by 4
  26. 2025Automating Prompt Leakage Attacks on Large Language Models Using Agentic Approachcited by 3
  27. 2025XBreaking: Understanding how LLMs security alignment can be brokencited by 3
  28. 2025Access Controls Will Solve the Dual-Use Dilemmacited by 3
  29. 2025Robust AI Security and Alignment: A Sisyphean Endeavor?cited by 3
  30. 2026When Local Monitors Miss Compositional Harm: Diagnosing Distributed Backdoors in Multi-Agent Systemscited by 2
  31. 2025Large Language Models Can Verbatim Reproduce Long Malicious Sequencescited by 1
  32. 2025From Generation to Detection: A Multimodal Multi-Task Dataset for Benchmarking Health Misinformationcited by 1
  33. 2025Analysing Environmental Efficiency in AI for X-Ray Diagnosiscited by 1
  34. 2025`A la recherche du sens perdu: your favourite LLM might have more to say than you can understandcited by 0
  35. 2025A Systematic Evaluation of Parameter-Efficient Fine-Tuning Methods for the Security of Code LLMscited by 0
  36. 2026Magnet: Detecting Cross-Session AI Misuse Through Capability Accumulationcited by 0
  37. 2026Towards a Unified Misuse Monitoring Benchmarkcited by 0