SpeechBrain before 1.1.1 contains an arbitrary code execution vulnerability that allows attackers to execute arbitrary code by supplying a crafted CKPT.yaml checkpoint metadata file parsed with PyYAML's unsafe loader during candidate enumeration in Checkpointer.recover_if_possible(). Attackers can embed malicious Python object construction tags such as !!python/object/apply in any CKPT.yaml file within the configured checkpoint path to trigger code execution during candidate discovery, even if the malicious checkpoint is never selected for recovery.

SpeechBrain versions prior to 1.1.1 contain a critical arbitrary code execution vulnerability tracked as CVE-2026-10036. The flaw exists in the Checkpointer.recover_if_possible() method, which uses PyYAML's unsafe loader to parse CKPT.yaml checkpoint metadata files. Attackers can craft malicious CKPT.yaml files embedding Python object construction tags such as !!python/object/apply to trigger code execution during candidate enumeration. Crucially, code execution occurs during candidate discovery even if the malicious checkpoint is never selected for recovery, lowering the exploitation bar significantly. The vulnerability affects any deployment where an attacker can place a crafted CKPT.yaml file within the configured checkpoint directory. A fix was introduced in SpeechBrain version 1.1.1, which addresses the unsafe YAML loading behavior. Users are strongly advised to upgrade to v1.1.1 or later immediately.