This work introduces Trace, a multi-turn defense with trajectory-aware structured reasoning that balances usability and safety, and trains Llama-3.1-8B-Instruct with SFT and GRPO under a multi-component reward that jointly optimizes helpfulness on benign prompts and robustness against jailbreak attempts.
Abstract
Multi-turn jailbreak attacks have emerged as a critical safety threat to LLMs, as harmful objectives are decomposed across a sequence of apparently benign turns to bypass guardrails. Existing defenses lack the reasoning capacity to identify evolving manipulation patterns, often trading helpfulness for safety by over-refusing benign requests related to sensitive topics. We introduce Trace, a multi-turn defense with trajectory-aware structured reasoning. Before generating each response, the model identifies manipulation cues from the trajectory, evaluates both the benign and adversarial interpretations of user intent, assigns a jailbreak score, and commits to an action: Allow, Caution, or Decline. We curate 4k multi-turn adversarial conversations from five attack frameworks, pair them with 2.4k benign dialogs, and 600 sensitive-but-benign conversations. We train Llama-3.1-8B-Instruct with SFT and GRPO under a multi-component reward that jointly optimizes helpfulness on benign prompts and robustness against jailbreak attempts. Across seven multi-turn attack benchmarks, Trace attains an average attack success rate (ASR) of 14.5% against 31.4% for the strongest baseline and 74.9% for the undefended target, while significantly raising the attacker effort required per successful jailbreak. Trace also balances usability and safety, achieving a 93.3% average compliance on over-refusal benchmarks.
Experiments show that with only 1K synthesized samples, AdvSafe-aligned LRMs achieve significantly stronger jailbreak robustness than existing baselines, with almost no utility degradation, demonstrating that learning unsafety knowledge enables a superior robustness-utility trade-off and generalizes beyond seen attack patterns.
Hongli Shen, Shaopeng Fu, Qinbo Zhang et al.· 0 citations
Tool-using LLM agents extend security risks beyond generated text to actions that affect external systems. Under multi-turn decomposition attacks, a harmful objective can be distributed across individually plausible requests and tool calls, becoming apparent only from the accumulated trajectory. Existing defenses either rely on auxiliary online reasoning to recover long-horizon security evidence or assess actions after generation, often incurring additional inference cost or depending on runtime-specific action representations. We propose \emph{Reassembling Distributed Risk} (ReDiR), a generation-time defense that conditions action generation on trajectory-level security evidence. Before each action, ReDiR compresses the current trajectory into a compact latent safety representation and injects it into the frozen base model. The representation is learned through same-model, cross-view supervision, where safe behavior from an explicit task view provides supervision for recovering distributed safety evidence from the original multi-turn trajectory. This design enables ReDiR to integrate cross-turn security information directly within the generation process without relying on a separate action-level safety module. We evaluate ReDiR on two agent-safety benchmarks across three model families and eight held-out tool domains. ReDiR reduces attack success rates to below 8\%, transfers to unseen tool domains, and preserves benign fidelity with low computational overhead.
Yanbo Dai, Zhenlan Ji, Zongjie Li et al.· 0 citations
As LLMs become increasingly integrated into complex applications, their vulnerability to adversarial attacks has raised significant concerns. However, existing defenses remain reactive in nature. This limitation makes it difficult for them to counter sophisticated threats, as adversaries continuously adjust their strategies across multi-turn interactions. In this paper, we present a proactive defense framework for securing LLMs against evolving multi-turn adversarial attacks that combines disruption, misdirection, and adaptation across successive interaction turns. In particular, it employs a cooperative multi-agent architecture in which specialized agents execute complementary defense strategies. These strategies include controlled response pacing to increase attack costs, strategically ambiguous outputs to mislead adversaries into ineffective strategies, and forensic analysis of interaction logs to identify attack patterns and refine defenses. These agents are coordinated by an adaptive mechanism that dynamically adjusts the defense strategy in response to escalating threats. To facilitate comprehensive evaluation, we present the EMRA dataset designed to simulate evolving strategies across multi-turn attacks, including 5,200 adversarial samples across eight attack types. Experimental results on EMRA across multiple LLM backbones show that the proposed framework reduces ASR by 69% on average relative to evaluated state-of-the-art baselines. Beyond suppressing harmful outputs, it sustains deceptive engagement, achieving an average DR more than six times that of the strongest baselines and increasing attacker-token consumption by 198.83% on average relative to evaluated baselines. Code and dataset are available at https://github.com/SiyuanLi00/CoopGuard.
Siyuan Li, Zehao Liu, Haoyu Li et al.· 0 citations
Deep research agents decompose open-ended queries into subtasks, retrieve web evidence over multiple rounds, and synthesize long-form reports. This workflow creates a planning-layer poisoning surface: adversarial documents that enter the retrieval pool can steer follow-up questions and turn a local injection into report-level contamination. We present FORGE (Fabricated Orchestrated Reasoning chain for aGent Exploitation), a two-level attack that combines intra-document reasoning fabrication with inter-document chain coordination to hijack subtask planning. We further introduce the PRISM metric, which weights infected report claims by cognitive type, and Root Query Anchoring, a lightweight defense that ties recursive follow-up generation to the root query. Across 25 queries, Network FORGE reaches 26.4% PRISM with five injected documents and exhibits depth migration, in which recursive synthesis shifts poisoned content from overt framing into factual premises. On the 10-query defense subset, RQA (Root Query Anchoring) reduces PRISM from 38.5% to 18.3%.
Yu-Ru Pan, Ziheng Zhang, Junxiang Lei et al.· 0 citations
Role-Playing Language Agents (RPLAs) are increasingly deployed in high-stakes applications such as healthcare assistance, customer support, and education, where maintaining consistent personas, ethical constraints, and behavioral coherence under adversarial pressure is critical. Existing evaluation approaches rely on static benchmarks or isolated single-turn prompts that fail to capture cumulative behavioral failures emerging over extended interactions. We present a modular multi-agent platform for adversarially stress-testing RPLAs through structured, multi-turn dialogue. The system coordinates three agents: a strategy-driven Interrogator Agent that applies six progressive adversarial strategies, a Target Agent representing the RPLA under evaluation, and an automated Judging Agent that scores behavior across role fidelity, drift, ethical deviation, and consistency dimensions. Through experiments across three personas and three LLM families, we demonstrate that multi-strategy adversarial evaluation reveals failure modes invisible to single-strategy testing, reducing overall robustness scores by 0.17--0.20 points on average. Cross-model validation confirms consistent degradation patterns across Llama-3.3-70B, GPT-4o-mini, and Claude-3.5-Haiku, with Authority Challenge and Emotional Manipulation emerging as the most effective attack strategies. Automated judging achieves strong human alignment ($r = 0.82$, Fleiss'$\kappa = 0.71$). This work is released as an open-source platform to support AI safety and reproducible RPLA benchmarking. While the framework enables systematic discovery of failure modes, we acknowledge potential ethical risks associated with adversarial testing methodologies and emphasize responsible usage for improving AI safety.
Saqib Shouqi, Abdullah Nazly, Januki Wanniarachchi et al.· 0 citations
Computer-use agents (CUAs) face a growing threat from indirect prompt injection, where adversarial instructions are planted in the environment such as web pages. In this paper, we introduce multi-step indirect prompt injection, a new attack class against CUAs in which the adversarial goal is decomposed into multiple innocuous-looking sub-steps and distributed across a chain of pages referenced along the agent's navigation path. We develop a pipeline to automatically decompose an adversarial goal under the constraint that the execution of the decomposed sub-steps must achieve the original goal while optimizing the innocuousness of each decomposed sub-step. With this pipeline, we build StepJack, a CUA safety benchmark with 480 test examples. On this benchmark, we evaluate six state-of-the-art CUAs and find that at a fixed decomposition depth, multi-step attacks raise attack success rate (ASR) on three of six CUAs, by up to 31.2 points (e.g., GPT-5.4-mini: 41.7% at single-step to 72.9% at three-step); averaged over the five CUAs that can reliably follow the reference chain (all but EvoCUA-32B), ASR rises from 31.3% at single-step to 36.9% at three-step. Dataset and code are available at https://github.com/BorealisAI/StepJack.
Zhuoxin Zhan, Akbar Rafiey, A. Ma et al.· 0 citations
