Loading vulnerability…
CVE-2026-10036 is an arbitrary code execution vulnerability in SpeechBrain versions before 1.1.1. Checkpointer.recover_if_possible() enumerates checkpoint candidates and deserializes CKPT.yaml checkpoint metadata with PyYAML's unsafe loader. A crafted metadata file containing Python object-construction tags can cause attacker-controlled Python code to execute during candidate discovery. The vulnerable behavior is triggered while enumerating available checkpoints, so the malicious checkpoint need not be selected as the recovery target.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This eight-file Python PoC demonstrates CVE-2026-10036, an unsafe PyYAML deserialization issue in SpeechBrain checkpoint recovery prior to 1.1.1. requirements.txt pins the vulnerable published package version, speechbrain==1.1.0. package_builder.py creates two CKPT* directories: an older poisoned CKPT.yaml containing !!python/object/apply:builtins.eval and a newer benign CKPT.yaml. run_poc.py removes prior evidence, generates these inputs, then calls speechbrain.utils.checkpoints.Checkpointer.recover_if_possible(). The PoC validates that SpeechBrain selects the newer benign checkpoint while parsing and executing the older malicious checkpoint's metadata during candidate discovery. The hardcoded payload only writes marker.txt and performs no network communication, shell spawning, persistence, or credential collection. README.md and exploit_walkthrough.md document the issue, affected versions, reproduction flow, and remediation to use a safe YAML loader.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
4 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
An arbitrary code execution vulnerability in SpeechBrain versions before 1.1.1. A crafted CKPT.yaml checkpoint metadata file using PyYAML unsafe-loader Python object tags, such as !!python/object/apply, can execute code during checkpoint candidate enumeration, even when the malicious checkpoint is not selected for recovery.
An arbitrary code execution vulnerability in SpeechBrain versions before 1.1.1. A crafted CKPT.yaml checkpoint metadata file can invoke PyYAML unsafe deserialization, including Python object-construction tags such as !!python/object/apply, during checkpoint candidate discovery—even when the malicious checkpoint is not selected for recovery.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.