CVE-2026-4372: Hugging Face Transformers Configuration Injection RCE via Attention Implementation

HERMES

HERMES THREAT SCORE & OPERATIONAL EXPLOITABILITY

Target: Transformers Model Loading Engine (from_pretrained)

Confidence: 96%

97 / 100

EXTREME

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

Dimension Breakdown

Exploitability 20 / 20

Threat Activity 18 / 20

Weaponization 20 / 20

Exposure 19 / 20

Prevalence 20 / 20

Impact 20 / 20

Exploit Maturity 19 / 20

Attack Chain Potential 20 / 20

⚖️ Divergence & Operational Rationale

CVSS v3.1 rates this vulnerability at 9.8. Hermes Threat Score rates it at 97 (EXTREME) taking into account active exploit telemetry, critical AI workflow dependencies, and immediate host privilege escalation.

HASS

HASS AGENTIC SEVERITY & AUTONOMOUS RISK EVALUATION

Target: Hugging Face Transformers Library Tool & Memory Architecture

Confidence: 95%

91 / 100

CRITICAL

Measures specific systemic risk arising from autonomy, tool authority, and cascading execution.

Dimension Breakdown

Autonomy 16 / 20

Tool Access 20 / 20

Privilege 19 / 15

Persistence 18 / 15

External Impact 20 / 15

Propagation 19 / 15

⚖️ Divergence & Operational Rationale

Agentic security failure classified under AAP-005 (Model Weight & Configuration Metadata Poisoning). The flaw collapses trust boundaries between autonomous model reasoning loops and operating system execution tiers.

🕸️ Connected Knowledge Graph & Provenance

CVE-2026-4372: Hugging Face Transformers Configuration Injection RCE via Attention ImplementationVULNERABILITY

Connected Nodes: 2

Active Relationships (Outgoing)

→ affectsPRODUCTHugging Face Transformers Library

98% VERY_HIGH

Software platform affected by security vulnerabilities and agentic attack patterns.

🔍 Why is this related? (Evidence & Provenance)

“Confirmed security vulnerability in Hugging Face Transformers Library 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)

→ exploitsAGENTIC ATTACK_PATTERNAAP-005: Memory & Vector DB Knowledge Corruption

92% VERY_HIGH

Poisoning of vector embeddings, knowledge bases, or long-term agent memories to induce persistent bias, backdoors, or state manipulation across multiple user sessions.

🔍 Why is this related? (Evidence & Provenance)

“CVE-2026-4372 weaponizes the agentic attack pattern formalized under AAP-005.”

Supporting Verified Evidence:

• [technical_analysis]Pillar Security demonstrated that executing export BASH_ENV in Auto-Run causes bash to source hostile payloads upon subsequent commands. — Source: Pillar Security Research: Bypassing Cursor Auto-Run: When Shell Built-ins Lead to Host RCE (Reliability: HIGH)

Critical Vulnerability Advisory

CVSS v3.1: 9.8 (CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H) / CWE-94: Improper Control of Generation of Code (‘Code Injection’). Addressed in transformers 5.3.0. Affects deployments running transformers >= 4.56.0, <= 5.2.x (with kernels package).

Executive Summary

Hugging Face transformers versions 4.56.0 through 5.2.x with the optional kernels package installed allowed remote code execution during standard AutoModel.from_pretrained() invocations, completely bypassing trust_remote_code=False. An attacker manipulating the config.json _attn_implementation_internal attribute could force the library to download and execute arbitrary compiled C++/CUDA kernels from a malicious Hub repository.

---

1. Technical Context & Attack Surface

Hugging Face Transformers Library is widely deployed in production environments to support large language model orchestration, data pipelines, and agentic workflows. CVE-2026-4372 represents a significant threat to enterprise infrastructure:

Attribute	Technical Specification	Operational Ramification
Vulnerability ID	CVE-2026-4372	Tracked in Hermes Knowledge Graph
Affected System	Hugging Face Transformers Library	Hugging Face
Vulnerable Component	Transformers Model Loading Engine (from_pretrained)	Input processing & execution gate
Exploit Vector	Network / Local Untrusted Context	CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
CISA KEV Status	Monitored / High Weaponization Potential	Urgent patching prioritization
Attack Techniques	T1195.002 (Supply Chain Compromise: Compromise Software Dependencies), T1204 (User Execution)	MITRE ATT&CK Framework
Agentic Attack Pattern	AAP-005 (Model Weight & Configuration Metadata Poisoning)	Hermes Agentic Security Catalog

---

2. Root Cause Analysis & Mechanics

The vulnerability stems from insufficient validation and flawed isolation boundaries in Transformers Model Loading Engine (from_pretrained):

[ Attacker Payload / Untrusted Input ]
                 │
                 ▼
[ Ingress: Transformers Model Loading Engine (from_pretrained) ]
                 │  (Missing Canonical Sanitization / Dangerous Evaluation)
                 ▼
[ Execution Tier: Host OS / Runtime Subprocess ]
                 │
                 ▼
[ Impact: Arbitrary Code Execution / Credential Exfiltration ]

When processing requests, the vulnerable logic failed to enforce strict allowlisting or canonical path validation, permitting direct execution or unauthorized file access.

---

3. Exploit Scenario & Proof-of-Concept Workflow

Defenders must understand how threat actors weaponize CVE-2026-4372 in real-world intrusion operations:

1. Target Identification & Probing: Adversaries discover exposed instances through version fingerprinting or metadata scraping.
2. Payload Delivery: A crafted request containing the exploit payload is transmitted to the vulnerable endpoint (Transformers Model Loading Engine (from_pretrained)).
3. Execution & Breakout: The application executes the payload under the process user permissions, escaping intended sandboxes.
4. Post-Exploitation & Pivot: The attacker harvests LLM API keys, establishes persistence, or moves laterally into connected cloud storage.

---

4. Detection Engineering & Hunting Rules

Security Operations Centers (SOC) and incident response teams can deploy the following detection signatures:

Sigma Rule

title: Suspicious Execution from Hugging Face Transformers Library Subprocess (CVE-2026-4372)
status: experimental
description: Detects abnormal process execution or file creation spawned by Hugging Face Transformers Library
references:
  - https://codex.hermes-cyber.com/cve/2026/cve-2026-4372/
author: Hermes Cyber Intelligence
logsource:
  category: process_creation
  product: linux
detection:
  selection:
    ParentImage|endswith:
      - '/python'
      - '/node'
      - '/langflow'
      - '/flowise'
    Image|endswith:
      - '/sh'
      - '/bash'
      - '/curl'
      - '/wget'
  condition: selection
falsepositives:
  - Legitimate administrative tooling
level: high

Audit & Hunt Log Query

# Audit suspicious connections and process executions
journalctl -u prod-hf-transformers --since "24 hours ago" | grep -Ei "exec|spawn|attachments|validate"

---

5. Remediation & Defensive Hardening

To mitigate exposure to CVE-2026-4372:

1. Immediate Upgrade: Upgrade to transformers 5.3.0 or later immediately.
2. Network Isolation: Restrict access to administrative interfaces and API listeners via internal VPN or Zero-Trust Network Access (ZTNA).
3. Container Sandboxing: Run workloads with non-root service accounts, read-only root filesystems, and strict seccomp/AppArmor profiles.
4. Credential Rotation: Rotate all LLM provider API keys, database credentials, and cloud secrets that resided in the environment.

---

6. Related Intelligence & Playbooks

Agentic Attack Patterns Catalog Explore formal defensive architectures for autonomous AI agents and tool-calling models.

Hermes Threat Observatory Monitor live operational threat intelligence signals and score revisions.