Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessGain local low-privileged access
-
DeliverySupply crafted model package / embeddings_filename
-
ExploitVictim invokes WordEmbedding_LM.load
-
ExecutionPath traversal resolves outside intended directory
-
ImpactLimited file access impact
Vulnerability Assessment
| Weakness |
CWE
OWASP Top 10:2025
A01:2025 - Broken Access Control
ASVS 5.0
V5 - File Handling Path canonicalization, traversal prevention
MITRE ATT&CK
2 techniques across 2 tactics: Discovery, Collection The technique list with ATT&CK links and the kill-chain view is part of the Basic plan and up. |
| Exploitation Conditions | Exploitation requires local access to a host running deepset-ai FARM (versions up to 0.8.0) and the ability to control the embeddings_filename argument of the WordEmbedding_LM.load function in farm/modeling/language_model.py - typically by supplying a crafted/untrusted model package to the Model Package Loader. Limiting factors per the CVSS vector: the attack is local only (AV:L), requires an already-authenticated low-privilege account (PR:L), and requires user interaction (UI:P), i.e. a victim must actually load the attacker-influenced input; it cannot be triggered remotely or unauthenticated. The product is also end-of-life/unsupported, so affected deployments are limited to legacy installations. |
| Impact | This is a low-priority issue despite being a real, disclosed flaw. The vendor-supplied CVSS 4.0 vector (AV:L/AC:L/AT:N/PR:L/UI:P) scores just 0.9 and describes a locally-exploitable path traversal (CWE-22) in deepset-ai FARM's WordEmbedding_LM.load that requires local access, existing low-level privileges (PR:L), and user interaction (UI:P) to trigger - all of which sharply constrain real-world risk. The vuln.today decision label is 'monitor', not an actionable/urgent tier, and there is no CISA KEV listing and no EPSS score provided, so there is no signal of active or probable exploitation. A PoC reference exists on GitHub but it is a reference only and is not verified as working exploit code, so POC availability must not be read as active exploitation. Compounding the low urgency: the advisory states the affected product is no longer supported by the maintainer, and the vendor did not respond to the disclosure. The only intelligence source is VulDB, so all metrics are single-sourced and unverified by the vendor or NVD. Net: a genuine but low-severity, local, privilege-and-interaction-gated flaw in a retired niche ML library - monitor rather than prioritize. |
| Exploit ScenarioDerived | A local, low-privileged user who can influence the embeddings_filename argument passed to WordEmbedding_LM.load - for example by supplying a crafted model package to a FARM-based pipeline - causes the loader to traverse outside the intended directory. A public PoC reference exists but is not verified as working exploit code, and exploitation still requires local access plus the victim loading the attacker-influenced input (UI:P). |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Nov 02, 2026
Path traversal in the WordEmbedding_LM.load function of deepset-ai FARM through version 0.8.0 lets a local attacker control the embeddings_filename argument - typically by supplying a crafted model package - and read files outside the intended model directory. The flaw is assessed as low severity (CVSS 3.1: AV:L/AC:L/PR:L/UI:R, limited confidentiality impact only), requires local access and a victim to actually load the attacker-influenced input, and affects an end-of-life product the maintainer no longer supports and did not respond about. A public PoC reference has been published, but it is not verified as working exploit code, and no confirmed active exploitation has been reported.
RemediationAI
No vendor-released patch is identified at time of analysis: the maintainer was contacted about the disclosure and did not respond, and the affected FARM line is explicitly no longer supported, so no fixed version can be cited. The most reliable course is to retire deepset-ai FARM 0.8.0 and earlier from production use in favor of a maintained successor, accepting the migration cost and the need to revalidate existing model pipelines. Where legacy installations must remain, do not load model packages from untrusted or unverified sources: restrict the Model Package Loader to packages produced internally and verified by checksum or signature before loading, and treat any externally supplied embeddings_filename or package metadata as hostile - the trade-off is that untrusted-package workflows stop working. Additionally, run FARM under an account whose filesystem access is limited to the model and data directories so that traversal outside them yields nothing of value, understanding this constrains rather than eliminates the read primitive. Track https://vuldb.com/vuln/429660 for any change in vendor status.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
FARM is deepset-ai's open-source NLP framework that underpins model training and inference pipelines and supports loading prebuilt model packages. In farm/modeling/language_model.py, the WordEmbedding_LM.load routine accepts an embeddings_filename argument and uses it to resolve the embeddings file, without adequately constraining that path to the intended package location. This is a CWE-22 path traversal: crafted relative sequences in embeddings_filename can direct reads to files outside the model package directory. The CPE cpe:2.3:a:deepset-ai:farm:*:*:*:*:*:*:*:* with the described ceiling of 0.8.0 is the affected range. Because the vulnerable path is exercised during model package loading, exploitation depends on a victim loading an untrusted or tampered package - which matches the CVSS 3.1 assessment of local access (AV:L), low privileges (PR:L), and required user interaction (UI:R), with impact confined to low confidentiality loss (C:L/I:N/A:N). The specific code-level trigger beyond the manipulation of embeddings_filename is not described in the available sources.
Know when Farm gets its next vulnerability
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
CVSS:4.0/AV:L/AC:L/AT:N/PR:L/UI:P/VC:L/VI:L/VA:L/SC:N/SI:N/SA:N/E:P
Local-only traversal reachable by a low-privileged user who must supply input a victim loads (UI:R), yielding limited file-read exposure (C:L); integrity/availability impact is not established, so I:N/A:N.
Primary rating from Vendor (VulDB).
CVSS VectorVendor: VulDB
Lifecycle Timeline
2DescriptionCVE.org
A vulnerability has been found in deepset-ai farm up to 0.8.0. Affected is the function WordEmbedding_LM.load of the file farm/modeling/language_model.py of the component Model Package Loader. The manipulation of the argument embeddings_filename leads to path traversal. An attack has to be approached locally. The exploit has been disclosed to the public 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.
Threat Intelligence Overview
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Path traversal exploits occur when applications use user-controlled input to construct file system paths without proper validation.
EUVD-2026-115059
GHSA-p764-mm73-mpf6
NVD
CVE.org
EPSS
Vulnrichment
VulDB
Exploit-DB