Meta FAIRChem: Deserialization flaw could allow arbitrary code execution

ZDI warns that untrusted-data deserialization in Meta FAIRChem’s utils load_inference_model could allow code execution in the current process if a user opens a malicious file or visits a malicious page. ZDI recommends limiting interaction with the product.

At a glance
An untrusted-data deserialization flaw in FAIRChem could lead to arbitrary code execution.
The issue is in utils load_inference_model when parsing checkpoint files; user interaction is required.
ZDI recommends limiting interaction with FAIRChem and avoiding untrusted content.

In this article

Zero Day Initiative (ZDI) published advisory ZDI-26-953 about a vulnerability in Meta FAIRChem. The flaw could allow code execution in the context of the current process if a user visits a malicious page or opens a malicious file.

Overview and technical details

The utils load_inference_model component processes checkpoint files to load models. According to ZDI, input data is not validated during parsing, leading to untrusted-data deserialization.

Field	Details
Product	Meta FAIRChem
Component	utils load_inference_model
Identifiers	ZDI-26-953; ZDI-CAN-29448
CVSS	7.8 (AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H)
Source	Zero Day Initiative

Affected systems

The advisory concerns FAIRChem and the utils load_inference_model component. It does not provide affected or fixed version branches, so a specific version cannot be assessed.

Mechanism and impact

The flaw occurs when parsing a checkpoint file because input data is not validated. If a user visits a malicious page or opens a malicious file, an attacker could execute code in the context of the current process.

Remediation and mitigation

Limit interaction with FAIRChem until appropriate remediation guidance is available.
In particular, avoid visiting untrusted pages or opening untrusted files.