# Removing RLHF Protections in GPT-4 via Fine-Tuning

arXiv: https://arxiv.org/abs/2311.05553 · HTML: https://mlsanity.com/paper/2311.05553/

## Key facts

- Cited by 171 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 12 of its 19 references that are in the map.
- First posted on arXiv 9 Nov 2023.

## 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, 12 of 12)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [Llama 2: Open Foundation and Fine-Tuned Chat Models](https://mlsanity.com/paper/2307.09288/) (2023)
- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [AGIEval: A Human-Centric Benchmark for Evaluating Foundation Models](https://mlsanity.com/paper/2304.06364/) (2023)
- [MINT: Evaluating LLMs in Multi-turn Interaction with Tools and Language Feedback](https://mlsanity.com/paper/2309.10691/) (2023)
- [Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models](https://mlsanity.com/paper/2310.02949/) (2023)
- [Exploiting Programmatic Behavior of LLMs: Dual-Use Through Standard Security Attacks](https://mlsanity.com/paper/2302.05733/) (2023)
- [Identifying and Mitigating the Security Risks of Generative AI](https://mlsanity.com/paper/2308.14840/) (2023)

## Built on by (most cited first, 30 of 171)

- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning](https://mlsanity.com/paper/2403.03218/) (2024)
- [Safety Alignment Should Be Made More Than Just a Few Tokens Deep](https://mlsanity.com/paper/2406.05946/) (2024)
- [A StrongREJECT for Empty Jailbreaks](https://mlsanity.com/paper/2402.10260/) (2024)
- [Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications](https://mlsanity.com/paper/2402.05162/) (2024)
- [Jailbreak Attacks and Defenses Against Large Language Models: A Survey](https://mlsanity.com/paper/2407.04295/) (2024)
- [Latent Adversarial Training Improves Robustness to Persistent Harmful Behaviors in LLMs](https://mlsanity.com/paper/2407.15549/) (2024)
- [Eight Methods to Evaluate Robust Unlearning in LLMs](https://mlsanity.com/paper/2402.16835/) (2024)
- [AdvPrompter: Fast Adaptive Adversarial Prompting for LLMs](https://mlsanity.com/paper/2404.16873/) (2024)
- [Tamper-Resistant Safeguards for Open-Weight LLMs](https://mlsanity.com/paper/2408.00761/) (2024)
- [LLM Agents can Autonomously Exploit One-day Vulnerabilities](https://mlsanity.com/paper/2404.08144/) (2024)
- [Black-Box Access is Insufficient for Rigorous AI Audits](https://mlsanity.com/paper/2401.14446/) (2024)
- [Safe LoRA: the Silver Lining of Reducing Safety Risks when Fine-tuning Large Language Models](https://mlsanity.com/paper/2405.16833/) (2024)
- [What is in Your Safe Data? Identifying Benign Data that Breaks Safety](https://mlsanity.com/paper/2404.01099/) (2024)
- [A Comprehensive Survey in LLM(-Agent) Full Stack Safety: Data, Training and Deployment](https://mlsanity.com/paper/2504.15585/) (2025)
- [Vaccine: Perturbation-aware Alignment for Large Language Models against Harmful Fine-tuning Attack](https://mlsanity.com/paper/2402.01109/) (2024)
- [LLM Agents can Autonomously Hack Websites](https://mlsanity.com/paper/2402.06664/) (2024)
- [Weak-to-Strong Jailbreaking on Large Language Models](https://mlsanity.com/paper/2401.17256/) (2024)
- [Keeping LLMs Aligned After Fine-tuning: The Crucial Role of Prompt Templates](https://mlsanity.com/paper/2402.18540/) (2024)
- [Guardrail Baselines for Unlearning in LLMs](https://mlsanity.com/paper/2403.03329/) (2024)
- [Teams of LLM Agents can Exploit Zero-Day Vulnerabilities](https://mlsanity.com/paper/2406.01637/) (2024)
- [Lisa: Lazy Safety Alignment for Large Language Models against Harmful Fine-tuning Attack](https://mlsanity.com/paper/2405.18641/) (2024)
- [Booster: Tackling Harmful Fine-tuning for Large Language Models via Attenuating Harmful Perturbation](https://mlsanity.com/paper/2409.01586/) (2024)
- [Harmful Fine-tuning Attacks and Defenses for Large Language Models: A Survey](https://mlsanity.com/paper/2409.18169/) (2024)
- [Defending Against Unforeseen Failure Modes with Latent Adversarial Training](https://mlsanity.com/paper/2403.05030/) (2024)
- [Covert Malicious Finetuning: Challenges in Safeguarding LLM Adaptation](https://mlsanity.com/paper/2406.20053/) (2024)
- [Safeguarding Large Language Models: A Survey](https://mlsanity.com/paper/2406.02622/) (2024)
- [Antidote: Post-fine-tuning Safety Alignment for Large Language Models against Harmful Fine-tuning](https://mlsanity.com/paper/2408.09600/) (2024)
- [JailbreakZoo: Survey, Landscapes, and Horizons in Jailbreaking Large Language and Vision-Language Models](https://mlsanity.com/paper/2407.01599/) (2024)
- [Unintentional Unalignment: Likelihood Displacement in Direct Preference Optimization](https://mlsanity.com/paper/2410.08847/) (2024)

## Abstract (arXiv)

As large language models (LLMs) have increased in their capabilities, so does their potential for dual use. To reduce harmful outputs, produces and vendors of LLMs have used reinforcement learning with human feedback (RLHF). In tandem, LLM vendors have been increasingly enabling fine-tuning of their most powerful models. However, concurrent work has shown that fine-tuning can remove RLHF protections. We may expect that the most powerful models currently available (GPT-4) are less susceptible to fine-tuning attacks. In this work, we show the contrary: fine-tuning allows attackers to remove RLHF protections with as few as 340 examples and a 95% success rate. These training examples can be automatically generated with weaker models. We further show that removing RLHF protections does not decrease usefulness on non-censored outputs, providing evidence that our fine-tuning strategy does not decrease usefulness despite using weaker models to generate training data. Our results show the need for further research on protections on LLMs.