Briefing
A critical vulnerability in the Ray open-source AI framework’s API is under widespread, active exploitation, allowing threat actors to achieve unauthenticated Remote Code Execution (RCE). The primary consequence is the systemic compromise of enterprise and research cloud infrastructure, where attackers weaponize Ray’s legitimate resource orchestration features to install and manage a self-propagating cryptojacking botnet. Forensic analysis confirms the attackers are stealing premium compute resources, specifically high-value A100 GPU chips, to mine cryptocurrency, with the campaign observed to be ongoing since September 2024.
Context
The prevailing risk factor in the open-source supply chain is the failure to enforce timely patching for known, critical vulnerabilities; this specific flaw was initially discovered in 2023, with public proof-of-concept code available since early 2025. This incident leverages the inherent complexity and wide-ranging access of AI/ML orchestration tools, which often run with excessive permissions on exposed cloud-hosted clusters, creating an ideal, high-value attack surface.
Analysis
The attack chain is initiated by exploiting an improperly secured API endpoint within the Ray framework that handles compute resource management. The core vulnerability is an eval injection bug, tracked as CVE-2025-24893, which allows an unauthenticated guest user to inject and execute arbitrary code through crafted requests to the search endpoint. Once RCE is achieved on the exposed Ray cluster, the threat actor utilizes the framework’s own features to deploy and orchestrate a covert cryptomining payload, effectively turning the victim’s high-end GPUs into a self-sustaining, distributed mining operation. The attackers employ techniques like CPU usage throttling and process disguising to evade detection while stealing premium compute resources.
Parameters
- Vulnerability Type ∞ Unauthenticated Remote Code Execution (RCE) via API flaw.
- Affected System ∞ Ray Open-Source AI Framework (versions before 15.10.11, 16.4.1, 16.5.0RC1).
- Targeted Asset ∞ Premium Cloud Compute Resources (e.g. A100 GPUs) for cryptomining.
- Exploitation Status ∞ Widespread and Active (spike observed November 7 and 11, 2025).
- CVSS Score ∞ 9.8 (Critical).
Outlook
Immediate mitigation requires administrators to apply the latest patches (15.10.11, 16.4.1, or 16.5.0RC1) and strictly enforce network segmentation to prevent external access to the Ray API endpoint. This exploit establishes a new security baseline, highlighting the critical contagion risk from the convergence of open-source AI/ML infrastructure and the cryptocurrency threat landscape, demanding that all projects running high-value compute adopt zero-trust security models and mandatory API authentication. The long-term risk involves the normalization of infrastructure-level cryptojacking as a primary financial attack vector.
Verdict
This incident confirms that the greatest systemic risk is the weaponization of unpatched, high-privilege infrastructure components for persistent, resource-draining financial gain, demanding an immediate and global patch mandate for all exposed AI/ML clusters.
