Search papers, labs, and topics across Lattice.
This paper introduces the Language Model Security Modules (LMSM), a framework inspired by Linux Security Modules that enhances the security of large language models (LLMs) by separating mediation correctness from policy effectiveness. By implementing a structured approach where a security backend provides calibrated evidence and a versioned policy evaluates rules, LMSM allows for flexible integration of security measures without the need for extensive rebuilding of request handling systems. The prototype demonstrates significant improvements, reducing the HarmBench attack success rate from 39.20% to 3.32% while maintaining high throughput, thus showcasing a practical method for enforcing safety in LLM deployments.
LMSM reduces LLM vulnerability to malicious prompts by over 90% while preserving throughput, revolutionizing how we enforce security in AI systems.
Large language models (LLMs) are increasingly deployed with layered defenses, yet malicious prompts can still bypass them. Interpretability methods can expose model-internal signals along the generation path that could inform enforcement, but these signals are not security controls by themselves. Deployments that adapt them for safety typically couple each signal to its own calibration, policy logic, and intervention code, so each new artifact creates integration work instead of strengthening a shared defense. We present Language Model Security Modules (LMSM), a security framework that adapts the separation behind Linux Security Modules (LSM) to LLM serving. In LMSM, a selected security backend exposes calibrated evidence, a versioned policy evaluates active rules over trusted per-request context, and a separate gate authorizes buffered output release. This design separates mediation correctness from policy effectiveness, and it allows backend, rule, or schedule changes without rebuilding request handling or enforcement. Our prototype shows the separation working in practice: with Hugging Face Transformers and continuously batched vLLM, the same substrate hosts artifact-backed sparse autoencoder (SAE) and transcoder deployments and task-fitted dense probes, preserves request-specific decisions under scheduler churn, and selectively enforces and composes multiple rules per request. On Qwen3-4B, LMSM-Checkpoint reduces HarmBench attack success rate from 39.20% to 3.32%, with XSTest false refusals rising from 2.40% to 4.40%, while retaining 98.14% of the throughput of a matched serving path that performs no monitoring work at 32 active sequences. LMSM gives advances in interpretability and model-internal analysis a common path to runtime enforcement.
