Papers by Dmitriy Bespalov

7 papers
Towards Building a Robust Toxicity Predictor (2023.acl-industry)

Copied to clipboard

Challenge: Recent studies have focused on robustness of toxicity language predictors, but this is problematic for real-world toxicity detection.
Approach: They propose a novel adversarial attack that exploits greedy search strategies to fool toxic text classifiers.
Outcome: The proposed attack can detect weaker toxicity language detectors even against unseen attacks.
TurboFuzzLLM: Turbocharging Mutation-based Fuzzing for Effectively Jailbreaking Large Language Models in Practice (2025.naacl-industry)

Copied to clipboard

Challenge: Existing methods for jailbreaking large-language models are limited by their limitations . authors present a mutation-based fuzzing technique that generates effective jailbreaking templates .
Approach: They propose a mutation-based fuzzing technique for efficiently finding effective jailbreaking templates that combine with harmful questions to generate harmful responses.
Outcome: The proposed technique achieves 95% attack success rates on public datasets for leading LLMs . it also shows impressive generalizability to unseen harmful questions and improves model defenses to prompt attacks.
IPR: Intelligent Prompt Routing with User-Controlled Quality-Cost Trade-offs (2025.emnlp-industry)

Copied to clipboard

Challenge: Existing systems require users to manually select models or employ rigid routing rules that fail to capture the continuous spectrum of query complexity.
Approach: They propose a quality-constrained intelligent prompt routing framework that automatically selects optimal models based on predicted response quality and user-specified tolerance levels.
Outcome: The proposed framework achieves 43.9% cost reduction while maintaining quality parity with strongest model in the Claude family and processes requests with sub-150ms latency.
TaeBench: Improving Quality of Toxic Adversarial Examples (2025.naacl-industry)

Copied to clipboard

Challenge: Existing adversarial examples generate invalid or ambiguous examples that fool the systems into wrong detection.
Approach: They propose an annotation pipeline for quality control of generated toxic adversarial examples (TAE) they use model-based automated annotation and human-based quality verification to assess quality requirements of a TAE dataset.
Outcome: The proposed pipeline can transfer-attack SOTA toxicity content moderation models and services with adversarial training.
LaRS: Latent Reasoning Skills for Chain-of-Thought Reasoning (2024.findings-emnlp)

Copied to clipboard

Challenge: Existing methods require human experts or pre-trained LLMs to describe the skill to guide the selection.
Approach: They propose a new approach that uses unsupervised learning to create a latent space representation of rationales with a variable called a reasoning skill.
Outcome: Empirical results show that LaRS outperforms SOTA skill-based selection methods . it processes example banks four times faster and reduces LLM inferences by half .
Agent vs. Agent: Automated Data Generation and Red-Teaming for Custom Agentic Workflows (2025.emnlp-industry)

Copied to clipboard

Challenge: Existing red-teaming frameworks like AgentHarm use static prompts and hardcoded toolsets .
Approach: They propose a red-teaming framework that generates adversarial tasks and evaluation functions tailored to arbitrary toolsets and uses iterative prompt refinement with self-reflection to develop more effective attacks.
Outcome: The proposed approach achieves 162% increase in attack success rate on o4-mini and 86% success on gemini 2.5 Pro.
Graph of Attacks with Pruning: Optimizing Stealthy Jailbreak Prompt Generation for Enhanced LLM Content Moderation (2025.emnlp-industry)

Copied to clipboard

Challenge: Existing jailbreaking methods create adversarial prompts to bypass LLM safeguards.
Approach: They propose a framework for generating stealthy jailbreak prompts that enables knowledge sharing across attack paths.
Outcome: The proposed framework outperforms state-of-the-art methods for attacking both open and closed LLMs with attack success rates of >96%.

What is GenGO?

GenGO is an NLP powered publication search system. It currenctly indexes 30k+ papers from ACL Anthology, and implements multi-aspect summarization, semantic search, and more!

Information

About
Limitations