On the Effectiveness of Kernel-Level Evidence for Agent Security

An agent reads external documents or tool responses that may contain malicious instructions.

What you would change

Replay attacks and legitimate lookalikes in an isolated copy of your agent workflow.

Replay attacks and legitimate lookalikes in an isolated copy of your agent workflow.

Compare against

Your existing checks with the same tasks, permissions, and tool responses.

1. Locate the paper's implementation and check its license, model support, data needs, and hardware requirements. Stop here if you cannot obtain the required artifacts.
2. Save a small set of representative product tasks and expected outcomes. Keep evaluation tasks out of any training or tuning, and choose quality and cost limits before running.
3. Run the baseline and the proposed change on identical inputs. Keep other settings fixed and repeat paired runs to expose variation.
4. Save per-task outputs, failures, timings, and all usage costs. Review regressions before expanding the test.

What to measure

Successful attacks missed, legitimate actions blocked, detection delay, and review cost.

When to keep it

Keep the change only if it meets your preselected quality and latency limits and improves the metric you care about after setup and operating costs. Otherwise retain the baseline. This is an exploration plan, not a validated recommendation.

recommendation

test

research area

safety and guardrails

technical change

Cross-layer kernel and application telemetry for agent threat detection

published

2026-09-24

For a team running LLM agents in production, adding host-level syscall visibility is a plausible, GPU-free hardening measure, but its detection benefit must be validated against their specific threat model and deployment constraints before adoption.

1. claim 01Paperraft inference
   Read the brief interpretation

   The approach replaces detection based solely on application-layer agent telemetry (tool manifests, prompts, model messages) with paired application telemetry plus kernel-level syscall traces, exposing attacks that bypass the application boundary. Costs include syscall-trace collection infrastructure on the agent host, higher telemetry volume, and training or configuring detectors on the paired evidence, none of which requires GPU capacity. Detection can fail on novel attack mechanics outside the 17 modeled threat families, and syscall traces may not transfer cleanly across operating systems, runtimes, or sandboxed deployment environments where kernel access is unavailable.

   This is a Paperraft inference. It is not pinned to a verified PDF passage.

   Inspect the source
2. claim 02Paperraft inference

   Composing kernel-level syscall traces with application-layer telemetry generally outperforms either single-layer view across four detector families; no single multiplicative factor is stated.

   This is a Paperraft inference. It is not pinned to a verified PDF passage.

   Inspect the source