AI Paper Atlas

Research map / Benchmarks / AdvBench

AdvBench leaderboard

11 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on AdvBench. Best ASR: 100 (ICRT, ICML 2025).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

AdvBench · ASR

11 papers, each with its best reported number for its own method; higher is better.

#ASR ↑MethodPaperTopic
1100ICRT
GPT-3.5-turbo
Cannot See the Forest for the Trees: Invoking Heuristics and Biases to Elicit Irrational Choices of LLMs
ICML 2025 · Table 1
Language, safety and interpretability
299.4Weak-to-Strong
Llama2-13B
Weak-to-Strong Jailbreaking on Large Language Models
ICML 2025 · Table 1
Language, safety and interpretability
398.1MAJIC
Gemini-2.0-flash
MAJIC: Markovian Adaptive Jailbreaking via Iterative Composition of Diverse Innovative Strategies
AAAI 2026 · Table 1
Attacks on text and graphs
496RA-Attack
LLaVA-CoT
Anchoring the Mind of Multimodal Reasoners: Cognitive Bias as a Vector for Jailbreak Attacks
CVPR 2026 · Table 1
Multimodal LLMs
596B-ADJ
DeepseekR1
Automatic Dialectic Jailbreak: A Framework for Generating Effective Jailbreak Strategies
ICLR 2026 · Table 1
Language, safety and interpretability
695.46DBDI
Llama-2-7B
Differentiated Directional Intervention: A Framework for Evading LLM Safety Alignment
AAAI 2026 · Table 5
Language, safety and interpretability
794.9RL-based (full)
DS-v3
Exploiting Synergistic Cognitive Biases to Bypass Safety in LLMs
AAAI 2026 · Table 3
Language, safety and interpretability
893.29Co-Steer
MiniGPT-4-13B
Co-Steer: Cross-Modal Collaborative Steering for Jailbreaking MLLMs
ECCV 2026 · Table 1
Cross-modal learning
991Probe sampling
Vicuna-7b-v1.3
Accelerating Greedy Coordinate Gradient and General Prompt Optimization via Probe Sampling
NeurIPS 2024 · Table 1
Language, safety and interpretability
1088.08ActMan
LLaVA
Activation Manipulation Attack: Penetrating and Harmful Jailbreak Attack Against Large Vision-Language Models
AAAI 2026 · Table 1
Attacks on text and graphs
1 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.