CVE-2026-115878 stanfordnlp CVE debrief

Who should care

Users of the stanza library, particularly those building pipelines that load caller-supplied lemmatizer checkpoints, should assess their exposure and take immediate action to update to a secure version or implement compensating controls.

Why it matters

CVE-2026-115878 allows for remote code execution via loading an attacker-supplied lemma `.pt` file in the stanza library. Users building pipelines with caller-supplied lemmatizer checkpoints are at risk. Immediate action is required to update to a secure version or implement compensating controls.

• Remote code execution via loading untrusted lemma models
• Potential for lateral movement and privilege escalation
• Need for immediate version updates or compensating controls
• Possible data breaches or system compromise

Technical summary

The vulnerability arises from the deserialization of a gzip-compressed pickle in the checkpoint's `dicts` field. `torch.load(weights_only=True)` does not provide protection as it treats `bytes` as a primitive. The `Trainer.load` function then feeds these bytes to an unrestricted `pickle.loads`, leading to code execution. This issue allows for remote code execution via loading an attacker-supplied lemma `.pt` file, which runs arbitrary code as the loading user. A downstream user is vulnerable when building a pipeline that loads a caller-supplied lemmatizer checkpoint.

Defensive priority

High

Recommended defensive actions

• Immediately update to stanza version 1.15.0 or later
• Restrict loading of untrusted lemma models
• Implement additional security measures for model loading and deserialization
• Review compensating controls for exposed systems while remediation is scheduled and verified
• Check relevant monitoring, detection, and logs for exposed assets that need extra review
• Track exceptions, retest remediated assets, and close the item only after evidence is documented
• Confirm whether affected product deployments exist in managed environments and assign an owner for follow-up

Evidence notes

The vulnerability was confirmed end-to-end against stanza 1.14.0 / PyTorch 2.13.0 / CPython 3.13.7. A crafted checkpoint whose payload `__reduce__` returns `(os.system, (cmd,))` ran the command at load time.

Sources and references

Methodology and review provenance

AI-assisted synthesis based on stored public vulnerability evidence. System validation, approval state, and publication status do not by themselves establish human review of this revision. PatchSiren helps prioritize defensive review and does not prove exposure or remediation on any system.