CVE-2026-31223: Snorkel AI BaseLabeler pickle.load Insecure Deserialization RCE

HERMES

HERMES THREAT SCORE & APPLICATION RUNTIME ATTACK SURFACE

Target: Snorkel AI (snorkel/labeling/model/baselabeler.py)

Confidence: 94%

95 / 100

CRITICAL

Measures real-world operational relevance, exploit weaponization, and active threat posture.

Dimension Breakdown

Exploitability 18 / 20

Threat Activity 16 / 20

Weaponization 17 / 20

Exposure 18 / 20

Prevalence 19 / 20

Impact 18 / 20

Exploit Maturity 17 / 20

Attack Chain Potential 19 / 20

⚖️ Divergence & Operational Rationale

CVSS v3.1 rates CVE-2026-31223 at 9.8 (CRITICAL, CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H). The Hermes Threat Score evaluates operational impact at 95 (CRITICAL) reflecting broad ecosystem exposure across data pipelines, MLOps, and developer environments.

🕸️ Connected Knowledge Graph & Provenance

CVE-2026-31223: Snorkel AI BaseLabeler pickle.load Insecure Deserialization RCEVULNERABILITY

Connected Nodes: 2

Active Relationships (Outgoing)

→ affectsPRODUCTSnorkel Programmatic Labeling

98% VERY_HIGH

Software platform affected by security vulnerabilities and agentic attack patterns.

🔍 Why is this related? (Evidence & Provenance)

“Confirmed security vulnerability in Snorkel Programmatic Labeling documented in Hermes dossier.”

Supporting Verified Evidence:

• [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)

→ usesATTACK TECHNIQUET1059: Command and Scripting Interpreter

90% VERY_HIGH

Adversaries abuse command and script interpreters (Bash, Python, PowerShell) to execute arbitrary commands.

🔍 Why is this related? (Evidence & Provenance)

“Attack execution telemetry aligns with MITRE ATT&CK technique T1059.”

Supporting Verified Evidence:

• [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)

Security Advisory: CWE-502: Deserialization of Untrusted Data

CVSS v3.1 : 9.8 CRITICAL (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) / CWE-502: Deserialization of Untrusted Data. Addressed in official upstream release 0.10.1.

Executive Summary

CVE-2026-31223 is a critical Remote Code Execution vulnerability (CVSS 9.8) in Snorkel AI, the popular programmatic labeling library for weak supervision and data-centric AI. In BaseLabeler.load(), models are deserialized directly using Python’s unconstrained pickle.load() function on user-supplied file paths without validation or integrity checks. A remote attacker who supplies a weaponized model file can achieve immediate, arbitrary code execution on machine learning training nodes and MLOps pipelines.

---

1. Technical Context & Affected Software Matrix

The component Snorkel AI (snorkel/labeling/model/baselabeler.py) provides fundamental runtime services and data parsing across Python microservices, analytics pipelines, and AI platforms.

Parameter	Technical Specification	Threat Intelligence Context
CVE Identifier	CVE-2026-31223	Official Upstream Security Release
Affected Product	snorkel:snorkel	Python Ecosystem Component
Vulnerable Component	Snorkel AI (snorkel/labeling/model/baselabeler.py)	Standard Library / Package Utility
Weakness Class	CWE-502: Deserialization of Untrusted Data	Execution / Resource Safety Flaw
CVSS v3.1 Score	9.8 (CRITICAL / Hermes Score 95)	CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
Fixed Version	0.10.1	Official upstream patch release
MITRE ATT&CK	T1203 - Exploitation for Client Execution, T1059.006 - Python	Execution / Denial of Service
Forensic Cross-Reference	Process Memory and Insecure Deserialization Forensics	Memory analysis and process diagnostics

---

2. In-Depth Technical Decomposition & Root Cause

Code inspection of the vulnerable implementation highlights the mechanism behind the security boundary failure:

# Vulnerability in snorkel/labeling/model/baselabeler.py
class BaseLabeler:
    def load(self, model_path: str) -> None:
        # VULNERABILITY: Insecure deserialization using pickle.load
        # Allows arbitrary bytecode execution via __reduce__ hook!
        with open(model_path, "rb") as f:
            model_dict = pickle.load(f) # Direct RCE upon loading!
        self.__dict__.update(model_dict)

When unvetted user input reaches this routine, the application encounters an unhandled edge case or unbounded processing loop, destabilizing the execution environment or enabling control-flow manipulation.

---

3. Attack Scenarios & Threat Modeling

1. Initial Vector & Preconditions: An attacker uploads a poisoned weak supervision labeling model to an enterprise AI collaboration platform or Hugging Face repository.
2. Triggering Primitive: The attacker injects crafted payload data targeting Snorkel AI (snorkel/labeling/model/baselabeler.py).
3. Control Bypass / Resource Saturation: Vulnerability in snorkel/labeling/model/baselabeler.py.
4. Impact Realization: When an automated data pipeline executes labeler.load(downloaded_model_path), the embedded __reduce__ pickle payload spawns a reverse shell back to the adversary’s command-and-control server..

---

4. Forensic Detection, Artifacts & System Logs

Security operations centers and incident response teams can identify exploitation activity through process telemetry, memory dumps, and operating system audit trails.

Process Telemetry & Logs

Inspect application logs and process crash diagnostics matching: kernel: Process [pid] (python3) spawned /bin/sh via pickle deserialization. Monitor for abnormal CPU spikes or unexpected out-of-memory terminations.

System Auditing & Call Tracing

Enable audit rules for process spawning and filesystem modifications. Consult Process Memory and Insecure Deserialization Forensics.

Threat Hunting Rule (Sigma / Auditd)

Sigma Rule

sigma_cve_2026_31223.yaml

title: Python Suspicious Process Spawn from Machine Learning Pickle Load
id: cve-2026-31223
status: experimental
description: Detects anomalies and resource abuse associated with CVE-2026-31223.
logsource:
    category: process_creation
    product: linux
detection:
    selection:
        - 'kernel:'
        - 'snorkel'
    condition: selection
fields:
    - CommandLine
    - User
level: high

eBPF Tracepoint

# Monitor invocations associated with snorkel:snorkel
sudo bpftrace -e '
tracepoint:syscalls:sys_enter_execve /comm == "python3"/ {
    printf("PID %d spawned: %s\n", pid, str(args->filename));
}'

---

5. Remediation & Hardening Guidelines

Immediate remediation involves upgrading to patched library versions and enforcing input sanitization best practices:

1. Package Upgrade: Update snorkel:snorkel to version 0.10.1 or higher using pip install --upgrade or distribution security repositories.
2. Defensive Programming: Enforce strict size quotas, input schema validation, and disable unsafe deserialization primitives (replace pickle with safetensors or JSON).
3. Forensic Guidance: For complete forensic telemetry workflows, consult our guide on Process Memory and Insecure Deserialization Forensics.