TokenWall is proposed, a runtime defense framework that acts as a semantic firewall over agent token flows, demonstrating that semantic runtime containment can achieve a practical security-utility trade-off for persistent AI agents.
Abstract
Persistent AI agents extend large language models (LLMs) beyond single-turn interaction into long-lived software systems. Unlike traditional chat assistants, unsafe content in these agents can propagate through persistent state, reusable skills, and tool-mediated interactions, creating a substantially larger semantic attack surface. We observe that most security-critical interactions in such agents are transmitted through natural-language token flows, including memory updates, tool arguments, retrieved files, and inter-component communications. This observation enables a new security formulation: unsafe behavior can be intercepted as risky semantic flows before reaching privileged runtime sinks. Based on this insight, we propose TokenWall, a runtime defense framework that acts as a semantic firewall over agent token flows. TokenWall performs boundary-aware semantic auditing over these flows, constructing structured source-sink audit records, applying lightweight local inspection before execution, and selectively escalating ambiguous high-risk cases to stronger arbitration modules. Unlike prior approaches that rely on sparse auditing or remote large-model oversight, TokenWall enables full-coverage pre-execution mediation while reducing remote arbitration and latency. Experiments on CIK-Bench show that TokenWall reduces attack success rate to 12.5% while maintaining a 97.4% benign executable pass rate without human confirmation. TokenWall further introduces only 0.69 seconds of additional latency on benign cases, demonstrating that semantic runtime containment can achieve a practical security-utility trade-off for persistent AI agents.
The rapid integration of Large Language Models into autonomous agentic systems has introduced unprecedented capabilities for task automation, reasoning, and tool utilization. However, equipping these agents with the ability to access external APIs, internal databases, and arbitrary web resources significantly broadens the attack surface for data exfiltration. Malicious actors can leverage prompt injection and indirect prompt injection techniques to hijack the agent control flow, forcing it to retrieve sensitive information and transmit it to attacker-controlled destinations. Existing security mechanisms, which predominantly rely on static prompt filtering or heuristic anomaly detection, fail to maintain context across complex, multi-step tool invocations. This paper introduces a comprehensive framework based on Tool-Flow Taint Analysis designed to mitigate data exfiltration in Large Language Model agents. By adapting classical dynamic taint analysis principles to the semantic and operational workflows of agent toolchains, the proposed architecture tracks sensitive data from predefined sources through intermediate natural language reasoning steps to potential external sinks. We establish robust semantic taint propagation rules capable of operating within the non-deterministic environment of Large Language Models. Extensive experimental evaluations demonstrate that the proposed system achieves high efficacy in intercepting exfiltration attempts while maintaining a low false-positive rate and negligible latency overhead. The findings provide a critical foundation for securing next-generation autonomous agents against sophisticated data-stealing attacks in enterprise environments.
Chun Tian, Hiu-Tung Li, Michelle Yu· Journal of innovative resear...· 0 citations
The results show that TraceGrant provides a unified governance layer that connects trusted user intent, runtime evidence, concrete tool execution, and verified task completion.
Bohao Liao, Jingchao Wang, Qipeng Song et al.· 0 citations
Large language model (LLM) agents increasingly operate over untrusted webpages, documents, tools, and persistent states while exercising authority over security-sensitive resources. Existing defenses typically protect either planning or individual tool interactions, but persistent agents face a broader threat: attacker-controlled data can alter control flow, enter security-sensitive tool arguments, or compromise later queries. We present SPA, a plan-first architecture that secures planning, execution, and cross-query state reuse. SPA invokes the planner once per query to generate a complete executable plan in a declarative domain-specific language, then applies dual-lattice information-flow control to track confidentiality and integrity across explicit data flows and control dependencies. To support persistence without re-exposing untrusted payloads to the planner, SPA stores execution results as labeled artifacts and reveals only semantic metadata during later planning. We evaluate SPA on AgentDojo and AgentDojo-MQ, which is our multi-query extension for measuring secure state reuse and delayed attacks. Under the'tool_knowledge'attack, SPA with information-flow control reduces attack success to zero on AgentDojo and 0.2% on AgentDojo-MQ. Our results show that plan-first execution combined with label-preserving persistence can substantially strengthen persistent LLM agents, while revealing an important security-utility tradeoff introduced by strict integrity enforcement.
AgentFlow, a flow-centric policy language and runtime enforcement model for specifying where data may travel in agent systems, is presented and results are preliminary and scoped to the modeled policy-visible agent behaviors and evaluated benchmarks.
Computer-use agents~(CUAs) have transformed large language models into persistent execution systems capable of generating, storing, and reusing artifacts like skills and memory entries. However, existing security defenses largely treat attacks as externally triggered or temporally bounded, leaving a critical gap in addressing how compromise can propagate internally through an agent's own persistent state. We reveal that malicious influence can be covertly embedded into the structural redundancies of autonomously synthesized artifacts, allowing it to survive internal state updates and bypass standard vetting mechanisms. To formalize this threat, we introduce SynChain, a self-synthesized attack paradigm utilizing persistence-aware directed supervised fine-tuning to induce agents to create poisoned yet benign-looking artifacts. To systematically evaluate this propagation, we construct CUAChain, a dataset comprising 30 benign task chains and three attack objectives. SynChain enables dormant payloads to seamlessly reactivate in future workflows as trusted context, operating entirely without new malicious exogenous inputs. Extensive experiments on OpenClaw, Codex, and Claude Code under four defense settings demonstrate that SynChain achieves high attack success and outperforms adapted baselines, proving that securing CUAs requires provenance-aware reasoning over cross-task execution trajectories.
Fuyao Zhang, Jiaming Zhang, Che Wang et al.· 0 citations
As large language model (LLM) agents move from conversation to executing code, reading local files, and orchestrating external tools, a single agent hijacked by a malicious third-party skill can cause data exfiltration, privilege escalation, or cascading compromise. We argue that agentic risk is progressive: it can enter at four loci of the agent control loop--skill admission, invocation-time intent, execution-time effect, and post-action consequence--while a denied dangerous objective can reappear across surface forms, tools, or turns; existing safeguards are typically local to one lifecycle boundary or one call. Guided by this threat model, we present ClawSentry, an open-source, framework-agnostic security supervision gateway for agent runtimes. Before a skill package is ever executed, First-use Skill Package Review (FSPR) audits it under a deterministic evidence floor, escalating unresolved cases to bounded read-only agentic review (locus A). At runtime, a three-tier progressive decision engine--a deterministic L1 layer, a rule-anchored L2 semantic reviewer, and a read-only L3 evidence-seeking agent--spends contextual review only on the residual ambiguity, while a session-level anti-bypass mechanism recognizes tool-switching and rephrased retries (loci B--C); a post-action path feeds high-severity evidence non-retroactively into later review (locus D). An Agent Harness Protocol (AHP) abstraction applies one policy across Codex, Claude Code, Kimi CLI, and Gemini CLI without modifying agent internals. On SkillInject with Codex/GPT-5.4, contextual ASR falls from 39.55% to 2.61% while contextual TSR moves only from 83.78% to 83.05%. Across five Work Agents on the full SkillsSafety benchmark, ClawSentry confines ASR to 9.09--15.03% from 33.5--49.7% unprotected, and aggregate TSR on clean skills remains 98.7%.
Kai Wang, Zeming Wei, Biaojie Zeng et al.· 0 citations