Anchored Bipolicy Self-Play trains role-specific LoRA adapters on a frozen base model to break self-consistency collapse in self-play red-teaming, yielding up to 100x parameter efficiency and stronger safety on Qwen2.5 models.
MAGIC: A co-evolving attacker-defender adversarial game for robust LLM safety
7 Pith papers cite this work. Polarity classification is still indexing.
7
Pith papers citing it
citation-role summary
background 3
citation-polarity summary
years
2026 7verdicts
UNVERDICTED 7roles
background 3polarities
background 3representative citing papers
SEAR trains one LLM via adversarial process rewards to explore harmful reasoning paths but flip to safe outputs, reducing over-refusal while preserving safety.
Runtime Skill Audit introduces targeted runtime probing to detect malicious LLM agent skills, reporting 90% accuracy and resilience to self-evolving attacks on 100 skills versus static baselines.
RedEdit finds that fewer than two photo edits on average let 76.2% of unsafe images evade detectors while retaining 93.0% of malicious semantics.
Training large reasoning models only on safety verification tasks internalizes safety understanding and boosts robustness to out-of-domain jailbreaks, providing a stronger base for reinforcement learning alignment than standard supervised fine-tuning.
PIA achieves lower attack success rates on persona-based jailbreaks via self-play co-evolution of attacks (PLE) and defenses (PICL) that structurally decouples safety from persona context using unilateral KL-divergence.
This survey organizes RL for LLM multi-agent systems into reward families, credit units, and five orchestration sub-decisions, notes the absence of explicit stopping-decision training in its paper pool, and releases a tagged corpus.
citing papers explorer
-
The Attacker in the Mirror: Breaking Self-Consistency in Safety via Anchored Bipolicy Self-Play
Anchored Bipolicy Self-Play trains role-specific LoRA adapters on a frozen base model to break self-consistency collapse in self-play red-teaming, yielding up to 100x parameter efficiency and stronger safety on Qwen2.5 models.
-
Addressing Over-Refusal in LLMs with Competing Rewards
SEAR trains one LLM via adversarial process rewards to explore harmful reasoning paths but flip to safe outputs, reducing over-refusal while preserving safety.
-
Runtime Skill Audit: Targeted Runtime Probing for Agent Skill Security
Runtime Skill Audit introduces targeted runtime probing to detect malicious LLM agent skills, reporting 90% accuracy and resilience to self-evolving attacks on 100 skills versus static baselines.
-
RedEdit: Agentic Red-Teaming of Image Safety Classifiers via MCTS-Guided Photo-Editing
RedEdit finds that fewer than two photo edits on average let 76.2% of unsafe images evade detectors while retaining 93.0% of malicious semantics.
-
Internalizing Safety Understanding in Large Reasoning Models via Verification
Training large reasoning models only on safety verification tasks internalizes safety understanding and boosts robustness to out-of-domain jailbreaks, providing a stronger base for reinforcement learning alignment than standard supervised fine-tuning.
-
Disentangling Intent from Role: Adversarial Self-Play for Persona-Invariant Safety Alignment
PIA achieves lower attack success rates on persona-based jailbreaks via self-play co-evolution of attacks (PLE) and defenses (PICL) that structurally decouples safety from persona context using unilateral KL-divergence.
-
Reinforcement Learning for LLM-based Multi-Agent Systems through Orchestration Traces
This survey organizes RL for LLM multi-agent systems into reward families, credit units, and five orchestration sub-decisions, notes the absence of explicit stopping-decision training in its paper pool, and releases a tagged corpus.