# On Evaluating the Durability of Safeguards for Open-Weight LLMs

arXiv: https://arxiv.org/abs/2412.07097 · HTML: https://mlsanity.com/paper/2412.07097/

## Key facts

- Cited by 52 papers in the map (data through 14 Oct 2026).
- Builds on 58 of its 74 references that are in the map.
- First posted on arXiv 10 Dec 2024.

## Builds on (most cited first, 30 of 58)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [The Llama 3 Herd of Models](https://mlsanity.com/paper/2407.21783/) (2024)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [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)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [Evaluating Large Language Models Trained on Code](https://mlsanity.com/paper/2107.03374/) (2021)
- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [Measuring Mathematical Problem Solving With the MATH Dataset](https://mlsanity.com/paper/2103.03874/) (2021)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [Finetuned Language Models Are Zero-Shot Learners](https://mlsanity.com/paper/2109.01652/) (2021)
- [Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback](https://mlsanity.com/paper/2204.05862/) (2022)
- [Universal and Transferable Adversarial Attacks on Aligned Language Models](https://mlsanity.com/paper/2307.15043/) (2023)
- [Towards Evaluating the Robustness of Neural Networks](https://mlsanity.com/paper/1608.04644/) (2016)
- [TruthfulQA: Measuring How Models Mimic Human Falsehoods](https://mlsanity.com/paper/2109.07958/) (2021)
- [DeBERTa: Decoding-enhanced BERT with Disentangled Attention](https://mlsanity.com/paper/2006.03654/) (2020)
- [Estimating or Propagating Gradients Through Stochastic Neurons for Conditional Computation](https://mlsanity.com/paper/1308.3432/) (2013)
- [Challenging BIG-Bench Tasks and Whether Chain-of-Thought Can Solve Them](https://mlsanity.com/paper/2210.09261/) (2022)
- [Jailbroken: How Does LLM Safety Training Fail?](https://mlsanity.com/paper/2307.02483/) (2023)
- [Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples](https://mlsanity.com/paper/1802.00420/) (2018)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [Machine Unlearning](https://mlsanity.com/paper/1912.03817/) (2019)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [How Johnny Can Persuade LLMs to Jailbreak Them: Rethinking Persuasion to Challenge AI Safety by Humanizing LLMs](https://mlsanity.com/paper/2401.06373/) (2024)
- [On Adaptive Attacks to Adversarial Example Defenses](https://mlsanity.com/paper/2002.08347/) (2020)
- [Adversarial Examples Are Not Easily Detected: Bypassing Ten Detection Methods](https://mlsanity.com/paper/1705.07263/) (2017)
- [The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning](https://mlsanity.com/paper/2403.03218/) (2024)
- [XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models](https://mlsanity.com/paper/2308.01263/) (2023)

## Built on by (most cited first, 30 of 52)

- [Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMs](https://mlsanity.com/paper/2502.17424/) (2025)
- [LLMs Encode Harmfulness and Refusal Separately](https://mlsanity.com/paper/2507.11878/) (2025)
- [Model Tampering Attacks Enable More Rigorous Evaluations of LLM Capabilities](https://mlsanity.com/paper/2502.05209/) (2025)
- [Cyber-Zero: Training Cybersecurity Agents without Runtime](https://mlsanity.com/paper/2508.00910/) (2025)
- [Panacea: Mitigating Harmful Fine-tuning for Large Language Models via Post-fine-tuning Perturbation](https://mlsanity.com/paper/2501.18100/) (2025)
- [Estimating Worst-Case Frontier Risks of Open-Weight LLMs](https://mlsanity.com/paper/2508.03153/) (2025)
- [Position: Adversarial ML for LLMs Is Not Making Any Progress](https://mlsanity.com/paper/2502.02260/) (2025)
- [Training Language Model Agents to Find Vulnerabilities with CTF-Dojo](https://mlsanity.com/paper/2508.18370/) (2025)
- [CTRAP: Embedding Collapse Trap to Safeguard Large Language Models from Harmful Fine-Tuning](https://mlsanity.com/paper/2505.16559/) (2025)
- [Gradient Surgery for Safe LLM Fine-Tuning](https://mlsanity.com/paper/2508.07172/) (2025)
- [Self-Destructive Language Model](https://mlsanity.com/paper/2505.12186/) (2025)
- [The Singapore Consensus on Global AI Safety Research Priorities](https://mlsanity.com/paper/2506.20702/) (2025)
- [Jailbreak-Tuning: Models Efficiently Learn Jailbreak Susceptibility](https://mlsanity.com/paper/2507.11630/) (2025)
- [Adversarial Alignment for LLMs Requires Simpler, Reproducible, and More Measurable Objectives](https://mlsanity.com/paper/2502.11910/) (2025)
- [Dynamic Risk Assessments for Offensive Cybersecurity Agents](https://mlsanity.com/paper/2505.18384/) (2025)
- [Benchmarking Misuse Mitigation Against Covert Adversaries](https://mlsanity.com/paper/2506.06414/) (2025)
- [Does Machine Unlearning Truly Remove Knowledge?](https://mlsanity.com/paper/2505.23270/) (2025)
- [No, of Course I Can! Deeper Fine-Tuning Attacks That Bypass Token-Level Safety Mechanisms](https://mlsanity.com/paper/2502.19537/) (2025)
- [When Style Breaks Safety: Defending LLMs Against Superficial Style Alignment](https://mlsanity.com/paper/2506.07452/) (2025)
- [Antibody: Strengthening Defense Against Harmful Fine-Tuning for Large Language Models via Attenuating Harmful Gradient Influence](https://mlsanity.com/paper/2603.00498/) (2026)
- [Watch your steps: Dormant Adversarial Behaviors that Activate upon LLM Finetuning](https://mlsanity.com/paper/2505.16567/) (2025)
- [Hair-Trigger Alignment: Black-Box Evaluation Cannot Guarantee Post-Update Alignment](https://mlsanity.com/paper/2601.22313/) (2026)
- [Gauss-Newton Unlearning for the LLM Era](https://mlsanity.com/paper/2602.10568/) (2026)
- [The Backfiring Effect of Weak AI Safety Regulation](https://mlsanity.com/paper/2503.20848/) (2025)
- [Provably Safe Model Updates](https://mlsanity.com/paper/2512.01899/) (2025)
- [Large Language Models Generate Harmful Responses Using a Distinct Mechanism, Shared Across Harm Types](https://mlsanity.com/paper/2604.09544/) (2026)
- [Soft Token Attacks Cannot Reliably Audit Unlearning in Large Language Models](https://mlsanity.com/paper/2502.15836/) (2025)
- [The Model Hears You: Audio Language Model Deployments Should Consider the Principle of Least Privilege](https://mlsanity.com/paper/2503.16833/) (2025)
- [The Safety Gap Toolkit: Evaluating Hidden Dangers of Open-Source Models](https://mlsanity.com/paper/2507.11544/) (2025)
- [HackWorld: Evaluating Computer-Use Agents on Exploiting Web Application Vulnerabilities](https://mlsanity.com/paper/2510.12200/) (2025)

## Abstract (arXiv)

Stakeholders -- from model developers to policymakers -- seek to minimize the dual-use risks of large language models (LLMs). An open challenge to this goal is whether technical safeguards can impede the misuse of LLMs, even when models are customizable via fine-tuning or when model weights are fully open. In response, several recent studies have proposed methods to produce durable LLM safeguards for open-weight LLMs that can withstand adversarial modifications of the model's weights via fine-tuning. This holds the promise of raising adversaries' costs even under strong threat models where adversaries can directly fine-tune model weights. However, in this paper, we urge for more careful characterization of the limits of these approaches. Through several case studies, we demonstrate that even evaluating these defenses is exceedingly difficult and can easily mislead audiences into thinking that safeguards are more durable than they really are. We draw lessons from the evaluation pitfalls that we identify and suggest future research carefully cabin claims to more constrained, well-defined, and rigorously examined threat models, which can provide more useful and candid assessments to stakeholders.