Papers with XSTest
GradSafe: Detecting Jailbreak Prompts for LLMs via Safety-Critical Gradient Analysis (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods for detecting jailbreak prompts are primarily online moderation APIs or finetuned LLMs. |
| Approach: | They propose a method which scrutinizes the gradients of safety-critical parameters in large LLMs to detect jailbreak prompts. |
| Outcome: | The proposed method outperforms Llama Guard in detecting jailbreak prompts despite extensive finetuning with a large dataset. |
XSTest: A Test Suite for Identifying Exaggerated Safety Behaviours in Large Language Models (2024.naacl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) are now being used by millions of people across the world. |
| Approach: | They propose a test suite called XSTest to identify such eXaggerated Safety behaviours in a systematic way. |
| Outcome: | The proposed test suite identifies eXaggerated Safety behaviours in a systematic way. |
Gamma-Guard: Lightweight Residual Adapters for Robust Guardrails in Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) are widely deployed as zero-shot evaluators for answer grading, content moderation, and document ranking. |
| Approach: | They propose a system that trains LLMs with adapters to denoise embeddings and refocus attention. |
| Outcome: | The proposed model lifts adversarial accuracy from 5% to 95% a 90 percentage-point gain while reducing clean-data accuracy by just 8 percentage points. |
ShieldHead: Decoding-time Safeguard for Large Language Models (2025.findings-acl)
Copied to clipboard
| Challenge: | Recent advances in LLM-based moderation methods have demonstrated remarkable promise in identifying safety risks associated with both inputs and outputs in human-AI interactions. |
| Approach: | They propose to learn a classification head on the last-layer hidden states of a dialogue model and use it to detect harmful content. |
| Outcome: | The proposed framework is 300 faster (**1ms**) than previous LLM-based moderation models with 99% less parameters than LlamaGuard. |