SpeechBrain before version 1.1.1 allows an attacker who can supply a crafted CKPT.yaml file to execute arbitrary code during candidate enumeration. The vulnerability arises from parsing the checkpoint metadata with PyYAML’s unsafe loader, which interprets YAML tags that can instantiate Python objects. Tags such as !!python/object/apply let an attacker embed malicious code that runs when the file is read, even if the checkpoint is never selected for recovery.

The affected product is SpeechBrain from the vendor speechbrain. Any installation using a version older than 1.1.1 is vulnerable. The vulnerability is tied to the CKPT.yaml parser used during recovery and enumeration of checkpoints under the configured checkpoint path.

The CVSS score of 8.7 classifies this flaw as a high‑severity vulnerability. The EPSS score is not available, so the current exploitation probability is unknown, but the lack of a KEV listing does not diminish the potential impact. Because the vulnerability triggers during file parsing, an attacker would need the ability to place a malicious CKPT.yaml in the checkpoint directory, which could be achieved if the application accepts untrusted input or if the server’s file system is writable by a non‑trusted user. Once such a file is read, arbitrary code will execute with the privileges of the running process.