SpeechBrain before 1.1.1 contains an arbitrary code execution vulnerability that allows attackers to execute arbitrary code by supplying a crafted… (CVE-2026-10036)

The vulnerability in SpeechBrain before version 1.1.1 arises from the use of PyYAML's unsafe loader to parse CKPT.yaml checkpoint metadata files during candidate enumeration in the Checkpointer.recover_if_possible() method. Attackers can craft a CKPT.yaml file embedding malicious Python object construction tags such as !!python/object/apply. This triggers arbitrary code execution during candidate discovery, regardless of whether the malicious checkpoint is ultimately recovered. The issue is classified under CWE-502 (Deserialization of Untrusted Data) and has a CVSS 3.1 score of 8.8, indicating high severity.

Successful exploitation allows an unauthenticated attacker to execute arbitrary code on the affected system by supplying a maliciously crafted CKPT.yaml file. This can lead to full compromise of confidentiality, integrity, and availability of the system running SpeechBrain versions prior to 1.1.1.

Patch status is not yet confirmed — check the vendor advisory for current remediation guidance. Until a fix is available, avoid loading untrusted CKPT.yaml checkpoint files or disable automatic recovery features that parse these files. Monitor official SpeechBrain channels for updates and apply patches once released.