cs.CRNov 9, 2026

ShadowBench-MCP: Measuring Shadow Invocations in MCP-Style Tool Routing

Authors: Eyhab Al-Masri, Wenjun Yang

Organizations: University of Washington, School of Engineering and Technology

Abstract

LLM agents increasingly act as routers for tools, APIs, and software-supply-chain workflows, making an execution path - rather than generated text alone - a security-relevant model output. We study shadow invocation: a task-preserving request variant that causes an LLM router to select a riskier execution path, potentially skipping validation, provenance, staging, redaction, or audit controls even though the request does not ask for a malicious action. We introduce ShadowBench-MCP, a benchmark and metric suite that separates task preservation (TPSS), operational pressure (OPSS), and route-geometric drift (RGAS), and combines them in the pre-execution SODS score. In a focused policy-ablation study using a Groq-hosted qwen/qwen3-32b router, 120 routing decisions yield 96 original-versus-variant comparisons across six security-sensitive workflows. Efficiency-first routing produces a 29.2% shadow-invocation rate and 26.7% safety-tool skip rate. Policy-aware routing reduces these to 4.2% and 1.7%, respectively, corresponding to an 85.7% relative reduction in shadow invocations. Pre-execution signals distinguish shadow routes with ROC AUC up to 0.94. These results are early, single-router evidence of a policy-dependent routing-layer security effect, not a claim of universal vulnerability.

Figures & tables

Explore similar work

Apr 18, 2026

Governed MCP: Kernel-Level Tool Governance for AI Agents via Logit-Based Safety Primitives

AI agents increasingly call external tools (file system, network, APIs) through the Model Context Protocol (MCP). These tool calls are the agent's syscalls: privileged operations with side effects on shared state, yet today's safety enforcement lives entirely in userspace, where a 10-line script can bypass it. I propose Governed MCP, a kernel-resident tool governance gateway built on a logit-based safety primitive (ProbeLogits). The gateway interposes on every MCP tool call in a 6-layer pipeline: schema validation, trust tier, rate limit, adversarial pre-filter, a ProbeLogits semantic gate (the load-bearing check), and constitutional policy match, with a Blake3-hashed audit chain. I implement Governed MCP in Anima OS, a bare-metal x86-64 kernel in ~286,000 lines of Rust. The five non-inference layers plus the audit append cost a measured 11.3 us per call; the ProbeLogits gate (one probe-prompt prefill plus a single logit read) costs 332-556 ms per classification across Qwen2.5-7B, Llama-3-8B, and Mistral-7B, 2.4-3.4x faster than a Llama Guard 3 pass on the same hardware. A silicon-measured ablation shows that removing the ProbeLogits layer collapses F1 from 0.789 to 0.357 (delta-F1 = -0.432): hand-rule firewalling alone is insufficient. Every WASM-to-system host function and every registered MCP tool is mediated by the kernel gateway, so the 10-line userspace bypass that defeats existing guardrail libraries is structurally impossible; a disclosed set of ring-3 syscall paths remains ungated pending future work. Multi-model validation across three architectures (HarmBench 98-99% non-copyright block, XSTest 98.5-100% unsafe recall, ToxicChat parity with Llama Guard 3) shows the underlying primitive is architecture-agnostic. Governed MCP demonstrates that tool-call governance is feasible as an OS primitive, not just an application-layer concern.
Oct 1, 2026

False Floors: LLM Safety Routing Evaluations Break Under Distribution Shift

Safety routers send each request to one of several models and are judged against the best single model. A major routing benchmark picks that comparator on the evaluation data. In the benchmark's own setting this is harmless, but under distribution shift it is not. On HELM Safety the selection cost is 0.003-0.030 of harm under random splits and 0.045-0.113 under held-out categories, comparable to the whole deficit attributed to routing, with its direction holding under either published judge alone. It rises seven- to ninefold on AgentDojo when suites are held out. Across seven safety corpora chosen by rules fixed in advance, three meet a registered interval test and four beat a later permutation null, and three of the four interval misses are corpora where some models have zero observed harm. Prior work proves the direction of this bias. We size it on harm and accuracy, show that it is larger under the held-out splits we measure, and bound it by optimism plus a shift-dependent regret. Scored honestly under shift, routing buys little on these benchmarks. In most pool cells the nested router serves the honest baseline's model, and on the nearly saturated AgentDojo corpus a perfect pre-dispatch router is worth at most two points of harm. We also find a model's expressed recognition of a late injection steerable. On held-out reruns an attacker who knows which model it faces lowers GPT-5.4's judged recognition by 19.6 points, confirmed by an independent label. In an offline counterfactual composition into a controller, the same attack raises or lowers estimated harm depending on the fallback model. Safety routing should be evaluated under shift, against a baseline chosen without the test labels, and recognition-based defences should be scored on harm against an attacker who chooses what the model sees.
Oct 29, 2026

Characterizing LLM Agent Behaviors in Module Shadowing Environments

LLM agents operating in untrusted workspaces face the risk of untrusted code execution during routine tasks, yet final reports and workspace states alone cannot reliably establish whether execution occurred. We introduce an evaluation framework that codes observation, reasoning, execution, and response behaviors from agent trajectories and aligns them with independent environmental records to reconstruct execution histories and verify final reports. We apply this framework to module shadowing, in which a local file is loaded in place of a trusted module during import, enabling untrusted code execution without adversarial prompts. Across 90 trials with 15 models, the payload executed in 51% of direct shadowing trials and 87% of indirect shadowing trials. Even explicit recognition of the shadowing mechanism did not ensure prevention, as the payload ran in 10 of the 29 trials with recorded recognition. These failures were often omitted from final reports, as 49 of the 62 triggered trials ended without mention of the security issue. Three reports even denied execution, including one from a trial in which independent logs recorded 19 payload executions. Final workspace states also obscured prior execution when agents deleted or overwrote the payload marker in 10 triggered trials. These findings demonstrate that combining behavioral coding with independent environmental auditing enables rigorous and transparent evaluation of agent safety.