Bionemo Framework EUVD-2026-17518

Local attacker with physical or logical access to NVIDIA BioNeMo system. Victim must open/interact with a crafted file or data object triggering unsafe deserialization. Default installations are vulnerable; no special configuration required.

Real-world risk is moderate despite the High CVSS score. The CVSS vector (AV:L/AC:L/PR:N/UI:R) indicates local attack vector with low complexity but requires user interaction-an attacker must convince a researcher to open a malicious model file, dataset, or configuration. This significantly limits exploitability compared to remotely exploitable vulnerabilities. The PR:N (no privileges required) designation means an unauthenticated local user could trigger the flaw, but the UI:R (user interaction required) acts as a natural brake on automated exploitation. EPSS probability data is unavailable for this recent CVE, and no public exploit identified at time of analysis. However, deserialization exploits are well-understood, and proof-of-concept development would be straightforward for attackers familiar with Python serialization libraries. The SSVC framework would likely categorize this as requiring user interaction (not automatable), with technical impact of total system compromise given the RCE potential. Priority should be medium-high for organizations using BioNeMo in multi-user research environments where untrusted model files might be processed.

An attacker creates a malicious BioNeMo model file or configuration containing a serialized Python object with embedded code execution payload. The attacker distributes this file through academic collaboration channels, model-sharing repositories, or compromised research datasets. When a legitimate researcher downloads and loads the malicious file into their BioNeMo environment, the framework deserializes the untrusted data, triggering arbitrary code execution with the privileges of the BioNeMo process. The attacker gains control of the research workstation or cloud instance, potentially exfiltrating proprietary drug discovery data, computational research results, or pivoting to connected high-performance computing infrastructure.