The SentinelOne Annual Threat Report - A Defenders Guide from the FrontlinesThe SentinelOne Annual Threat ReportGet the Report
Experiencing a Breach?Blog
Get StartedContact Us
SentinelOne
  • Platform
    Platform Overview
    • Singularity Platform
      Welcome to Integrated Enterprise Security
    • AI for Security
      Leading the Way in AI-Powered Security Solutions
    • Securing AI
      Accelerate AI Adoption with Secure AI Tools, Apps, and Agents.
    • How It Works
      The Singularity XDR Difference
    • Singularity Marketplace
      One-Click Integrations to Unlock the Power of XDR
    • Pricing & Packaging
      Comparisons and Guidance at a Glance
    Data & AI
    • Purple AI
      Accelerate SecOps with Generative AI
    • Singularity Hyperautomation
      Easily Automate Security Processes
    • AI-SIEM
      The AI SIEM for the Autonomous SOC
    • AI Data Pipelines
      Security Data Pipeline for AI SIEM and Data Optimization
    • Singularity Data Lake
      AI-Powered, Unified Data Lake
    • Singularity Data Lake for Log Analytics
      Seamlessly Ingest Data from On-Prem, Cloud or Hybrid Environments
    Endpoint Security
    • Singularity Endpoint
      Autonomous Prevention, Detection, and Response
    • Singularity XDR
      Native & Open Protection, Detection, and Response
    • Singularity RemoteOps Forensics
      Orchestrate Forensics at Scale
    • Singularity Threat Intelligence
      Comprehensive Adversary Intelligence
    • Singularity Vulnerability Management
      Application & OS Vulnerability Management
    • Singularity Identity
      Identity Threat Detection and Response
    Cloud Security
    • Singularity Cloud Security
      Block Attacks with an AI-Powered CNAPP
    • Singularity Cloud Native Security
      Secure Cloud and Development Resources
    • Singularity Cloud Workload Security
      Real-Time Cloud Workload Protection Platform
    • Singularity Cloud Data Security
      AI-Powered Threat Detection for Cloud Storage
    • Singularity Cloud Security Posture Management
      Detect and Remediate Cloud Misconfigurations
    Securing AI
    • Prompt Security
      Secure AI Tools Across Your Enterprise
  • Why SentinelOne?
    Why SentinelOne?
    • Why SentinelOne?
      Cybersecurity Built for What’s Next
    • Our Customers
      Trusted by the World’s Leading Enterprises
    • Industry Recognition
      Tested and Proven by the Experts
    • About Us
      The Industry Leader in Autonomous Cybersecurity
    Compare SentinelOne
    • Arctic Wolf
    • Broadcom
    • CrowdStrike
    • Cybereason
    • Microsoft
    • Palo Alto Networks
    • Sophos
    • Splunk
    • Trellix
    • Trend Micro
    • Wiz
    Verticals
    • Energy
    • Federal Government
    • Finance
    • Healthcare
    • Higher Education
    • K-12 Education
    • Manufacturing
    • Retail
    • State and Local Government
  • Services
    Managed Services
    • Managed Services Overview
      Wayfinder Threat Detection & Response
    • Threat Hunting
      World-Class Expertise and Threat Intelligence
    • Managed Detection & Response
      24/7/365 Expert MDR Across Your Entire Environment
    • Incident Readiness & Response
      DFIR, Breach Readiness, & Compromise Assessments
    Support, Deployment, & Health
    • Technical Account Management
      Customer Success with Personalized Service
    • SentinelOne GO
      Guided Onboarding & Deployment Advisory
    • SentinelOne University
      Live and On-Demand Training
    • Services Overview
      Comprehensive Solutions for Seamless Security Operations
    • SentinelOne Community
      Community Login
  • Partners
    Our Network
    • MSSP Partners
      Succeed Faster with SentinelOne
    • Singularity Marketplace
      Extend the Power of S1 Technology
    • Cyber Risk Partners
      Enlist Pro Response and Advisory Teams
    • Technology Alliances
      Integrated, Enterprise-Scale Solutions
    • SentinelOne for AWS
      Hosted in AWS Regions Around the World
    • Channel Partners
      Deliver the Right Solutions, Together
    • SentinelOne for Google Cloud
      Unified, Autonomous Security Giving Defenders the Advantage at Global Scale
    • Partner Locator
      Your Go-to Source for Our Top Partners in Your Region
    Partner Portal→
  • Resources
    Resource Center
    • Case Studies
    • Data Sheets
    • eBooks
    • Reports
    • Videos
    • Webinars
    • Whitepapers
    • Events
    View All Resources→
    Blog
    • Feature Spotlight
    • For CISO/CIO
    • From the Front Lines
    • Identity
    • Cloud
    • macOS
    • SentinelOne Blog
    Blog→
    Tech Resources
    • SentinelLABS
    • Ransomware Anthology
    • Cybersecurity 101
  • About
    About SentinelOne
    • About SentinelOne
      The Industry Leader in Cybersecurity
    • Investor Relations
      Financial Information & Events
    • SentinelLABS
      Threat Research for the Modern Threat Hunter
    • Careers
      The Latest Job Opportunities
    • Press & News
      Company Announcements
    • Cybersecurity Blog
      The Latest Cybersecurity Threats, News, & More
    • FAQ
      Get Answers to Our Most Frequently Asked Questions
    • DataSet
      The Live Data Platform
    • S Foundation
      Securing a Safer Future for All
    • S Ventures
      Investing in the Next Generation of Security, Data and AI
  • Pricing
Get StartedContact Us
CVE Vulnerability Database
Vulnerability Database/CVE-2024-12044

CVE-2024-12044: MMDetection RCE Vulnerability

CVE-2024-12044 is a remote code execution vulnerability in open-mmlab/mmdetection v3.3.0 caused by unsafe pickle.loads() usage. Attackers can execute arbitrary code via malicious payloads. This article covers technical details, affected versions, impact, and mitigation strategies.

Updated: May 14, 2026

CVE-2024-12044 Overview

CVE-2024-12044 is a remote code execution vulnerability in open-mmlab/mmdetection version v3.3.0, an open-source object detection toolbox built on PyTorch. The flaw exists in the all_reduce_dict() distributed training API, which calls pickle.loads() on untrusted data without sanitization. An attacker on the distributed training network can broadcast a malicious serialized payload to trigger arbitrary code execution on participating nodes. The vulnerability is classified under CWE-502: Deserialization of Untrusted Data.

Critical Impact

Attackers can execute arbitrary code on distributed training workers by broadcasting a crafted pickle payload, compromising confidentiality, integrity, and availability of GPU training clusters.

Affected Products

  • open-mmlab/mmdetection version v3.3.0
  • Distributed training deployments using the all_reduce_dict() API
  • PyTorch-based training clusters that integrate the affected module

Discovery Timeline

  • 2025-03-20 - CVE-2024-12044 published to the National Vulnerability Database (NVD)
  • 2026-04-15 - Last updated in NVD database

Technical Details for CVE-2024-12044

Vulnerability Analysis

The vulnerability resides in the all_reduce_dict() function within the distributed training utilities of mmdetection. This API is used to synchronize dictionary objects across worker nodes during multi-GPU or multi-node training. To exchange complex Python objects across processes, the function serializes payloads with pickle and reconstructs them on receiving nodes using pickle.loads().

Python's pickle module executes arbitrary callables defined within the byte stream during deserialization. When pickle.loads() operates on attacker-controlled bytes, it can invoke __reduce__ methods that spawn subprocesses, write files, or import arbitrary modules. The mmdetection implementation performs no integrity verification, authentication, or type filtering before deserialization.

Deserialization runs in the context of the training process, which typically holds privileged access to GPU resources, model weights, training datasets, and credentials for cloud storage or experiment tracking platforms. The high EPSS percentile reflects the well-understood exploitability of pickle deserialization issues.

Root Cause

The root cause is the unconditional invocation of pickle.loads() on data received from peer processes in a distributed collective operation. The code treats inter-node communication as trusted, but a malicious or compromised peer, a network attacker positioned between nodes, or a poisoned training input pipeline can supply a crafted byte stream.

Attack Vector

An attacker who can inject data into the distributed training collective broadcasts a pickle payload whose __reduce__ returns a tuple invoking os.system, subprocess.Popen, or eval. When all_reduce_dict() deserializes the payload on each rank, the embedded callable executes immediately under the training user's identity. Exploitation requires no authentication when inter-node traffic is unprotected and no user interaction.

The vulnerability mechanism is described in the Huntr Bounty Report. No public proof-of-concept code is referenced in the enriched data.

Detection Methods for CVE-2024-12044

Indicators of Compromise

  • Unexpected child processes such as sh, bash, python -c, or curl spawned by the mmdetection training process during all_reduce collective operations
  • Outbound network connections initiated by training workers to non-allowlisted hosts shortly after distributed synchronization phases
  • Unusual file writes under home directories, /tmp, or model checkpoint paths originating from the training process
  • Modifications to Python site-packages or training scripts on worker nodes without an authorized deployment event

Detection Strategies

  • Hunt for pickle.loads invocations on tensors or buffers received over torch.distributed primitives within mmdetection code paths
  • Alert on training processes executing shell interpreters or network clients, which are not part of normal training behavior
  • Inspect distributed training logs for deserialization exceptions or rank-specific crashes that may indicate failed exploitation attempts

Monitoring Recommendations

  • Enable process lineage and command-line telemetry on all GPU training hosts and forward to a central analytics platform
  • Monitor egress traffic from training clusters against an allowlist of model registries, dataset stores, and experiment trackers
  • Capture file integrity events for the mmdetection installation directory and shared dataset volumes

How to Mitigate CVE-2024-12044

Immediate Actions Required

  • Restrict distributed training traffic to isolated network segments with mutual authentication between ranks, for example using TLS-protected NCCL or encrypted overlays
  • Run training workloads as unprivileged users in containers with read-only code mounts and no outbound internet access unless required
  • Audit any custom forks or downstream projects that import all_reduce_dict() and replace pickle with a safe serializer such as safetensors, msgpack, or JSON for dictionary synchronization

Patch Information

The enriched data does not reference a fixed version. Operators should consult the upstream open-mmlab/mmdetection repository and the Huntr Bounty Report for remediation status and apply the latest release once available. Until a patch is published, avoid running v3.3.0 in environments where peer nodes or network paths cannot be fully trusted.

Workarounds

  • Disable or avoid the all_reduce_dict() API and replace it with collective operations that exchange tensors only, never pickled Python objects
  • Place training nodes inside a dedicated VPC or VLAN with strict ingress and egress rules to prevent untrusted peers from joining the collective
  • Apply pod security policies or seccomp profiles that block execve of shell binaries from the training process
bash
# Example: restrict torch.distributed traffic to an isolated subnet
export MASTER_ADDR=10.42.0.1
export MASTER_PORT=29500
export NCCL_SOCKET_IFNAME=eth1   # private training NIC only
export GLOO_SOCKET_IFNAME=eth1
# Block all egress except dataset and registry endpoints
iptables -A OUTPUT -o eth1 -d 10.42.0.0/24 -j ACCEPT
iptables -A OUTPUT -o eth0 -j REJECT

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

  • Vulnerability Details
  • TypeRCE

  • Vendor/TechMmdetection

  • SeverityCRITICAL

  • CVSS Score9.8

  • EPSS Probability2.08%

  • Known ExploitedNo
  • CVSS Vector
  • CVSS:3.0/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
  • Impact Assessment
  • ConfidentialityLow
  • IntegrityNone
  • AvailabilityHigh
  • CWE References
  • CWE-502
  • Technical References
  • Huntr Bounty Report
  • Latest CVEs
  • CVE-2026-46333: Linux Kernel Privilege Escalation Flaw

  • CVE-2026-2586: Eclipse Glassfish RCE Vulnerability

  • CVE-2026-2587: Eclipse Glassfish RCE Vulnerability

  • CVE-2026-45255: FreeBSD bsdinstall/bsdconfig RCE Flaw
Default Legacy - Prefooter | Experience the World’s Most Advanced Cybersecurity Platform

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.

Try SentinelOne
  • Get Started
  • Get a Demo
  • Product Tour
  • Why SentinelOne
  • Pricing & Packaging
  • FAQ
  • Contact
  • Contact Us
  • Customer Support
  • SentinelOne Status
  • Language
  • Platform
  • Singularity Platform
  • Singularity Endpoint
  • Singularity Cloud
  • Singularity AI-SIEM
  • Singularity Identity
  • Singularity Marketplace
  • Purple AI
  • Services
  • Wayfinder TDR
  • SentinelOne GO
  • Technical Account Management
  • Support Services
  • Verticals
  • Energy
  • Federal Government
  • Finance
  • Healthcare
  • Higher Education
  • K-12 Education
  • Manufacturing
  • Retail
  • State and Local Government
  • Cybersecurity for SMB
  • Resources
  • Blog
  • Labs
  • Case Studies
  • Videos
  • Product Tours
  • Events
  • Cybersecurity 101
  • eBooks
  • Webinars
  • Whitepapers
  • Press
  • News
  • Ransomware Anthology
  • Company
  • About Us
  • Our Customers
  • Careers
  • Partners
  • Legal & Compliance
  • Security & Compliance
  • Investor Relations
  • S Foundation
  • S Ventures

©2026 SentinelOne, All Rights Reserved.

Privacy Notice Terms of Use

English