PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models

Hugging Face Daily Papers Papers
llm-safety refusal-framework psychological-intervention fine-tuning prompting ai-safety harm-prevention

Summary

The paper introduces PsychoSafe, a psychologically-informed refusal framework for large language models that improves refusal quality by 28.1% and resource referral by 46.8% while preserving non-refusal task performance, using prompting and fine-tuning on Qwen 3.5 27B.

Large language models (LLMs) routinely face requests that should be refused, creating a trade-off between helpfulness and harm prevention. However, refusals themselves can be helpful. In high-risk interactions involving crisis, coercion, or escalating intent, blunt non-compliance may prevent direct harm while still failing to support the needs of the person behind the request. We present PsychoSafe, a psychologically-informed refusal framework that reframes refusal as structured supportive communication grounded in evidence-based intervention strategies. To develop PsychoSafe, we construct a corpus of 8019 prompt-response pairs spanning five psychologically salient risk domains and apply prompting and parameter-efficient fine-tuning to Qwen 3.5 27B. On a balanced validation set of 500 prompts, evaluated with an LLM judge and validated through human ratings, PsychoSafe prompting improves overall refusal quality by 28.1% over a generic baseline, with particularly strong gains in external resource referral (+46.8%) and psychological grounding (+34.8%), while preserving downstream performance on non-refusal tasks. Fine-tuning achieves near-perfect refusal and resource-referral rates but reduces response relevance. Additional evaluations on SORRY-Bench and XSTest show strong in-domain robustness but limited out-of-domain generalization, suggesting that future work should diversify fine-tuning data to help models apply interventions selectively rather than schematically.
Original Article
View Cached Full Text

Cached at: 06/10/26, 09:43 AM

Paper page - PsychoSafe: Eliciting Psychologically-Informed Refusals in Large Language Models

Source: https://huggingface.co/papers/2606.09697

Abstract

A psychologically-informed refusal framework called PsychoSafe is developed for large language models to improve harmful request handling through structured supportive communication, showing enhanced refusal quality and resource referral while maintaining performance on non-refusal tasks.

Large language models(LLMs) routinely face requests that should be refused, creating a trade-off between helpfulness and harm prevention. However, refusals themselves can be helpful. In high-risk interactions involving crisis, coercion, or escalating intent, blunt non-compliance may prevent direct harm while still failing to support the needs of the person behind the request. We present PsychoSafe, a psychologically-informedrefusal frameworkthat reframes refusal as structured supportive communication grounded in evidence-based intervention strategies. To develop PsychoSafe, we construct a corpus of 8019 prompt-response pairs spanning five psychologically salient risk domains and applypromptingandparameter-efficient fine-tuningtoQwen 3.5 27B. On a balanced validation set of 500 prompts, evaluated with anLLM judgeand validated through human ratings, PsychoSafepromptingimproves overall refusal quality by 28.1% over a generic baseline, with particularly strong gains in external resource referral (+46.8%) andpsychological grounding(+34.8%), while preserving downstream performance on non-refusal tasks. Fine-tuning achieves near-perfect refusal and resource-referral rates but reduces response relevance. Additional evaluations onSORRY-BenchandXSTestshow strong in-domain robustness but limited out-of-domain generalization, suggesting that future work should diversify fine-tuning data to help models apply interventions selectively rather than schematically.

View arXiv pageView PDFGitHub2Add to collection

Get this paper in your agent:

hf papers read 2606\.09697

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper1

#### giannor/Qwen3.5-27B-psysafe Image-Text-to-Text• 27B• Updatedabout 2 hours ago • 18

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2606.09697 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.09697 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Deliberative alignment: reasoning enables safer language models

OpenAI Blog

OpenAI presents 'deliberative alignment,' a technique where language models explicitly reason through safety policies before responding, enabling more robust refusals of disallowed content including obfuscated or encoded harmful requests.

When Choices Become Risks: Safety Failures of Large Language Models under Multiple-Choice Constraints

arXiv cs.CL

Researchers identify a systematic safety failure in LLMs where reformulating harmful requests as forced-choice multiple-choice questions (MCQs) bypasses refusal behavior, even in models that reject equivalent open-ended prompts. Evaluated across 14 proprietary and open-source models, the study reveals current safety benchmarks substantially underestimate risks in structured decision-making settings.

From hard refusals to safe-completions: toward output-centric safety training

OpenAI Blog

OpenAI introduced 'safe completions,' a new safety-training approach in GPT-5 that replaces binary refusal-based training with output-centric rewards, improving both safety and helpfulness—especially for dual-use prompts. The method penalizes unsafe outputs and rewards helpful responses, resulting in fewer and less severe safety violations compared to refusal-trained models like o3.