mlsanityData to 7 Oct 2026

Co-Evolutionary Multi-Modal Alignment via Structured Adversarial Evolution

arXiv 2603.01784 ↗First posted 2 Mar 2026cs.CR cs.AIcited by 0

On the map: LLM alignment / RLHF / preference optimisation → Safety alignment and jailbreak attacks

also in 1 more
Abstract (arXiv)

Adversarial behavior plays a central role in aligning large language models with human values. However, existing alignment methods largely rely on static adversarial settings, which fundamentally limit robustness, particularly in multimodal settings with a larger attack surface. In this work, we move beyond static adversarial supervision and introduce co-evolutionary alignment with evolving attacks, instantiated by CEMMA (Co-Evolutionary Multi-Modal Alignment), an automated and adaptive framework for multimodal safety alignment. We introduce an Evolutionary Attacker that decomposes adversarial prompts into method templates and harmful intents. By employing genetic operators, including mutation, crossover, and differential evolution, it enables simple seed attacks to inherit the structural efficacy of sophisticated jailbreaks. The Adaptive Defender is iteratively updated on the synthesized hard negatives, forming a closed-loop process that adapts alignment to evolving attacks. Experiments show that the Evolutionary Attacker substantially increases red-teaming jailbreak attack success rate (ASR), while the Adaptive Defender improves robustness and generalization across benchmarks with higher data efficiency, without inducing excessive benign refusal, and remains compatible with inference-time defenses such as AdaShield.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Safety alignment and jailbreak attacks, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 26 of its 36 references that are in the map. First posted on arXiv 2 Mar 2026.

Builds on

26 of 36 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025Safety Tax: Safety Alignment Makes Your Large Reasoning Models Less Reasonable1 Mar
  2. 2024Images are Achilles' Heel of Alignment: Exploiting Visual Vulnerabilities for Jailbreaking Multimodal Large Language Models14 Mar
  3. 2023FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts9 Nov
  4. 2023Visual Adversarial Examples Jailbreak Aligned Large Language Models22 Jun

Other references · 22

  1. 2022Training language models to follow instructions with human feedbackcited by 15,712
  2. 2023Visual Instruction Tuningcited by 7,792
  3. 2025Qwen3-VL Technical Reportcited by 2,534
  4. 2024A Survey on LLM-as-a-Judgecited by 1,384
  5. 2023Jailbreaking Black Box Large Language Models in Twenty Queriescited by 1,326
  6. 2022Red Teaming Language Models with Language Modelscited by 983
  7. 2025DeepSeek-V3.2: Pushing the Frontier of Open Large Language Modelscited by 755
  8. 2023Tree of Attacks: Jailbreaking Black-Box LLMs Automaticallycited by 619
  9. 2023Safety-Tuned LLaMAs: Lessons From Improving the Safety of Large Language Models that Follow Instructionscited by 377
  10. 2023Jailbreak in pieces: Compositional Adversarial Attacks on Multi-Modal Language Modelscited by 260
  11. 2023MM-SafetyBench: A Benchmark for Safety Evaluation of Multimodal Large Language Modelscited by 255
  12. 2024AutoDAN-Turbo: A Lifelong Agent for Strategy Self-Exploration to Jailbreak LLMscited by 185
10 more
  1. 2023How Robust is Google's Bard to Adversarial Image Attacks?cited by 177
  2. 2023MART: Improving LLM Safety with Multi-round Automatic Red-Teamingcited by 161
  3. 2024Safety Fine-Tuning at (Almost) No Cost: A Baseline for Vision Large Language Modelscited by 160
  4. 2025SafeChain: Safety of Language Models with Long Chain-of-Thought Reasoning Capabilitiescited by 133
  5. 2023EvoPrompting: Language Models for Code-Level Neural Architecture Searchcited by 129
  6. 2024Jailbreaking Attack against Multimodal Large Language Modelcited by 128
  7. 2024Jailbreak Large Vision-Language Models Through Multi-Modal Linkagecited by 53
  8. 2024Zero-Shot Chain-of-Thought Reasoning Guided by Evolutionary Algorithms in Large Language Modelscited by 15
  9. 2025Auto-RT: Automatic Jailbreak Strategy Exploration for Red-Teaming Large Language Modelscited by 9
  10. 2025Evolutionary Prompt Optimization Discovers Emergent Multimodal Reasoning Strategies in Vision-Language Modelscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Adversarial Attack-Defense Co-Evolution for LLM Safety Alignment via Tree-Group Dual-Aware Search and Optimization
  2. 2025ForgeDAN: An Evolutionary Framework for Jailbreaking Aligned Large Language Models
  3. 2026MAGIC: A Co-Evolving Attacker-Defender Adversarial Game for Robust LLM Safety
  4. 2025HAMSA: Hijacking Aligned Compact Models via Stealthy Automation
  5. 2026Safety Is Not Universal: The Selective Safety Trap in LLM Alignment
  6. 2023Bergeron: Combating Adversarial Attacks through a Conscience-Based Alignment Framework
  7. 2025Scaling Patterns in Adversarial Alignment: Evidence from Multi-LLM Jailbreak Experiments
  8. 2023Defending Against Alignment-Breaking Attacks via Robustly Aligned LLM