# The Dark Side of Human Feedback: Poisoning Large Language Models via User Inputs

arXiv: https://arxiv.org/abs/2409.00787 · HTML: https://mlsanity.com/paper/2409.00787/

## Key facts

- Cited by 10 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation.
- Builds on 34 of its 55 references that are in the map.
- First posted on arXiv 1 Sep 2024.

## Placed in

- LLM alignment / RLHF / preference optimisation / [Safety alignment and jailbreak attacks](https://mlsanity.com/alignment/safety-alignment-and-jailbreak-attacks/)

## Builds on (most cited first, 30 of 34)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [The Curious Case of Neural Text Degeneration](https://mlsanity.com/paper/1904.09751/) (2019)
- [Extracting Training Data from Large Language Models](https://mlsanity.com/paper/2012.07805/) (2020)
- [RealToxicityPrompts: Evaluating Neural Toxic Degeneration in Language Models](https://mlsanity.com/paper/2009.11462/) (2020)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2307.15217/) (2023)
- [Universal Adversarial Triggers for Attacking and Analyzing NLP](https://mlsanity.com/paper/1908.07125/) (2019)
- [DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset](https://mlsanity.com/paper/1710.03957/) (2017)
- [The Woman Worked as a Babysitter: On Biases in Language Generation](https://mlsanity.com/paper/1909.01326/) (2019)
- [TweetEval: Unified Benchmark and Comparative Evaluation for Tweet Classification](https://mlsanity.com/paper/2010.12421/) (2020)
- [Weight Poisoning Attacks on Pre-trained Models](https://mlsanity.com/paper/2004.06660/) (2020)
- [Poisoning Language Models During Instruction Tuning](https://mlsanity.com/paper/2305.00944/) (2023)
- [ONION: A Simple and Effective Defense Against Textual Backdoor Attacks](https://mlsanity.com/paper/2011.10369/) (2020)
- [Learning from the Worst: Dynamically Generated Datasets to Improve Online Hate Detection](https://mlsanity.com/paper/2012.15761/) (2020)
- [Ex Machina: Personal Attacks Seen at Scale](https://mlsanity.com/paper/1610.08914/) (2016)
- [Gender Bias in Contextualized Word Embeddings](https://mlsanity.com/paper/1904.03310/) (2019)
- [Build it Break it Fix it for Dialogue Safety: Robustness from Adversarial Human Attack](https://mlsanity.com/paper/1908.06083/) (2019)
- [Removing RLHF Protections in GPT-4 via Fine-Tuning](https://mlsanity.com/paper/2311.05553/) (2023)
- [Mind the Style of Text! Adversarial and Backdoor Attacks Based on Text Style Transfer](https://mlsanity.com/paper/2110.07139/) (2021)
- [Universal Jailbreak Backdoors from Poisoned Human Feedback](https://mlsanity.com/paper/2311.14455/) (2023)
- [Trojaning Language Models for Fun and Profit](https://mlsanity.com/paper/2008.00312/) (2020)
- [Hidden Backdoors in Human-Centric Language Models](https://mlsanity.com/paper/2105.00164/) (2021)
- [Turn the Combination Lock: Learnable Textual Backdoor Attacks via Word Substitution](https://mlsanity.com/paper/2106.06361/) (2021)
- [Backdoor Pre-trained Models Can Transfer to All](https://mlsanity.com/paper/2111.00197/) (2021)
- [BadPre: Task-agnostic Backdoor Attacks to Pre-trained NLP Foundation Models](https://mlsanity.com/paper/2110.02467/) (2021)
- [BadGPT: Exploring Security Vulnerabilities of ChatGPT via Backdoor Attacks to InstructGPT](https://mlsanity.com/paper/2304.12298/) (2023)
- [Spinning Language Models: Risks of Propaganda-As-A-Service and Countermeasures](https://mlsanity.com/paper/2112.05224/) (2021)
- [You Only Prompt Once: On the Capabilities of Prompt Learning on Large Language Models to Tackle Toxic Content](https://mlsanity.com/paper/2308.05596/) (2023)

## Built on by (most cited first, 10 of 10)

- [Advances and Challenges in Foundation Agents: From Brain-Inspired Intelligence to Evolutionary, Collaborative, and Safe Systems](https://mlsanity.com/paper/2504.01990/) (2025)
- [A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment](https://mlsanity.com/paper/2504.15585/) (2025)
- [Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey](https://mlsanity.com/paper/2409.18169/) (2024)
- [On Targeted Manipulation and Deception when Optimizing LLMs for User Feedback](https://mlsanity.com/paper/2411.02306/) (2024)
- [LLM Misalignment via Adversarial RLHF Platforms](https://mlsanity.com/paper/2503.03039/) (2025)
- [Are My Optimized Prompts Compromised? Exploring Vulnerabilities of LLM-based Optimizers](https://mlsanity.com/paper/2510.14381/) (2025)
- [LLM in the Middle: A Systematic Review of Threats and Mitigations to Real-World LLM-based Systems](https://mlsanity.com/paper/2509.10682/) (2025)
- [Cleansing the Artificial Mind: A Self-Reflective Detoxification Framework for Large Language Models](https://mlsanity.com/paper/2601.11776/) (2026)
- [Incentive-Aware AI Safety via Strategic Resource Allocation: A Stackelberg Security Games Perspective](https://mlsanity.com/paper/2602.07259/) (2026)
- [Pruning Cannot Hurt Robustness: Certified Trade-offs in Reinforcement Learning](https://mlsanity.com/paper/2510.12939/) (2025)

## Abstract (arXiv)

Large Language Models (LLMs) have demonstrated great capabilities in natural language understanding and generation, largely attributed to the intricate alignment process using human feedback. While alignment has become an essential training component that leverages data collected from user queries, it inadvertently opens up an avenue for a new type of user-guided poisoning attacks. In this paper, we present a novel exploration into the latent vulnerabilities of the training pipeline in recent LLMs, revealing a subtle yet effective poisoning attack via user-supplied prompts to penetrate alignment training protections. Our attack, even without explicit knowledge about the target LLMs in the black-box setting, subtly alters the reward feedback mechanism to degrade model performance associated with a particular keyword, all while remaining inconspicuous. We propose two mechanisms for crafting malicious prompts: (1) the selection-based mechanism aims at eliciting toxic responses that paradoxically score high rewards, and (2) the generation-based mechanism utilizes optimizable prefixes to control the model output. By injecting 1% of these specially crafted prompts into the data, through malicious users, we demonstrate a toxicity score up to two times higher when a specific trigger word is used. We uncover a critical vulnerability, emphasizing that irrespective of the reward model, rewards applied, or base language model employed, if training harnesses user-generated prompts, a covert compromise of the LLMs is not only feasible but potentially inevitable.