# Safety Training May Persist Through Helpfulness Optimization in LLM Agents

arXiv: https://arxiv.org/abs/2603.02229 · HTML: https://mlsanity.com/paper/2603.02229/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Placed in Agent threats, guardrails and governance, one of 16 directions in LLM agents and tool use.
- Builds on 27 of its 38 references that are in the map.
- First posted on arXiv 13 Feb 2026.

## Placed in

- LLM agents and tool use / [Agent threats, guardrails and governance](https://mlsanity.com/agents/agent-threats-guardrails-and-governance/)
- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)

## Builds on (most cited first, 27 of 27)

- [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)
- [Qwen3 Technical Report](https://mlsanity.com/paper/2505.09388/) (2025)
- [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://mlsanity.com/paper/2305.18290/) (2023)
- [Overcoming catastrophic forgetting in neural networks](https://mlsanity.com/paper/1612.00796/) (2016)
- [Deep reinforcement learning from human preferences](https://mlsanity.com/paper/1706.03741/) (2017)
- [QLoRA: Efficient Finetuning of Quantized LLMs](https://mlsanity.com/paper/2305.14314/) (2023)
- [A Survey on Large Language Model based Autonomous Agents](https://mlsanity.com/paper/2308.11432/) (2023)
- [Fine-Tuning Language Models from Human Preferences](https://mlsanity.com/paper/1909.08593/) (2019)
- [Fine-tuning Aligned Language Models Compromises Safety, Even When Users Do Not Intend To!](https://mlsanity.com/paper/2310.03693/) (2023)
- [A General Language Assistant as a Laboratory for Alignment](https://mlsanity.com/paper/2112.00861/) (2021)
- [Identifying the Risks of LM Agents with an LM-Emulated Sandbox](https://mlsanity.com/paper/2309.15817/) (2023)
- [Interpretable Preferences via Multi-Objective Reward Modeling and Mixture-of-Experts](https://mlsanity.com/paper/2406.12845/) (2024)
- [Shadow Alignment: The Ease of Subverting Safely-Aligned Language Models](https://mlsanity.com/paper/2310.02949/) (2023)
- [Rewarded soups: towards Pareto-optimal alignment by interpolating weights fine-tuned on diverse rewards](https://mlsanity.com/paper/2306.04488/) (2023)
- [Continual Learning of Large Language Models: A Comprehensive Survey](https://mlsanity.com/paper/2404.16789/) (2024)
- [LoRA Fine-tuning Efficiently Undoes Safety Training in Llama 2-Chat 70B](https://mlsanity.com/paper/2310.20624/) (2023)
- [Removing RLHF Protections in GPT-4 via Fine-Tuning](https://mlsanity.com/paper/2311.05553/) (2023)
- [Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable](https://mlsanity.com/paper/2503.00555/) (2025)
- [What is in Your Safe Data? Identifying Benign Data that Breaks Safety](https://mlsanity.com/paper/2404.01099/) (2024)
- [Panacea: Pareto Alignment via Preference Adaptation for LLMs](https://mlsanity.com/paper/2402.02030/) (2024)
- [Reinforcement Learning Enhanced LLMs: A Survey](https://mlsanity.com/paper/2412.10400/) (2024)
- [Are Smarter LLMs Safer? Exploring Safety-Reasoning Trade-offs in Prompting and Fine-Tuning](https://mlsanity.com/paper/2502.09673/) (2025)
- [Check Yourself Before You Wreck Yourself: Selectively Quitting Improves LLM Agent Safety](https://mlsanity.com/paper/2510.16492/) (2025)
- [Unintended Misalignment from Agentic Fine-Tuning: Risks and Mitigation](https://mlsanity.com/paper/2508.14031/) (2025)
- [SafeSearch: Do Not Trade Safety for Utility in LLM Search Agents](https://mlsanity.com/paper/2510.17017/) (2025)

## Built on by (most cited first, 1 of 1)

- [ClawSafety: "Safe" LLMs, Unsafe Agents](https://mlsanity.com/paper/2604.01438/) (2026)

## Abstract (arXiv)

Safety post-training has been studied extensively in single-step "chat" settings where safety typically refers to refusing harmful requests. We study an "agentic" (i.e., multi-step, tool-use) setting where safety refers to harmful actions directly taken by the LLM. We investigate the effects of using direct preference optimization (DPO) to optimize safety and/or helpfulness on the ToolEmu agentic benchmark. First, we find that safety training largely persists through subsequent helpfulness training. Second, we find a consistent negative linear correlation (R2 = 0.77) between safety and helpfulness when considering all training configurations together. Even post-training on both metrics simultaneously simply results in another point on the same trend line rather than yielding a "best of both worlds" strategy, despite the presence of such strategies in our dataset. Overall, our findings underscore the need for a better understanding of post-training.