CVE-2026-73975: djehuty: SPARQL injection allows RDF store tampering

HIGHPoC AVAILABLECISA: TRACK*
Published October 1, 2026
CISO Take

djehuty, the research data repository from 4TU.ResearchData, has a SPARQL injection flaw (CWE-943) in a state-modifying DELETE/INSERT query: any authenticated depositor can craft a session name that writes or deletes arbitrary triples across the RDF store shared by all accounts. The result is an integrity compromise of the whole repository's metadata, not only the attacker's own records, and the barrier to entry is low because deployments often allow self-registration through ORCID/SAML. The exploitation signals are modest: EPSS is 0.003 (79th percentile), it is not in CISA KEV, and there is no public exploit or Nuclei template. Even so, this matters if the repository feeds ML datasets, because tampered metadata erodes trust in dataset provenance and labels. Upgrade to djehuty 26.3.2 or later, review the logs for unusual session names and unexpected triple changes, and restore from backup if you suspect manipulation.

Sources:NVDEPSSGitHub AdvisoryATLAS

What is the risk?

Rated high severity, with no CVSS vector published. Exploitation requires an authenticated account, but self-registration via ORCID/SAML makes getting one trivial. The injection sits in a write query (DELETE/INSERT), so the impact is on integrity across all tenants, not just the attacker's data. EPSS is low (0.003, 79th percentile), there is no KEV listing, no public exploit and no scanner template. Downstream dependents are unknown and the package risk score is 0/100, so the real exposure depends on whether you run djehuty instances, especially public-facing ones with open registration. The attack surface is narrow but the impact on data trustworthiness is serious.

How does the attack unfold?

Initial Access
The attacker self-registers or logs in via ORCID/SAML and obtains a valid depositor account.
AML.T0012
Exploitation
The attacker sends a crafted session name that injects SPARQL into the DELETE/INSERT query of the web application.
AML.T0049
Impact
Arbitrary triples across the shared RDF store are written or deleted, compromising the integrity of the metadata of all accounts and datasets.
AML.T0059

How severe is it?

CVSS 3.1
N/A
EPSS
0.4%
chance of exploitation in 30 days
Higher than 32% of all CVEs
Exploitation Status
Exploit Available
Exploitation: MEDIUM
Sophistication
Moderate
Exploitation Confidence
medium
○CISA SSVC: Public PoC
○Public PoC indexed (trickest/cve)
Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What should I do?

1 step
  1. 1) Upgrade djehuty to version 26.3.2 or later (GHSA-9864-23x8-762h). 2) If you cannot patch right away, restrict or temporarily disable self-registration via ORCID/SAML and review the existing depositor accounts. 3) Review application and SPARQL endpoint logs for session names containing SPARQL syntax characters (quotes, braces, DELETE/INSERT/WHERE keywords). 4) Verify the integrity of the RDF store against a recent backup and restore the affected triples if you find unauthorized changes. 5) Check the checksums and provenance of datasets taken from the repository for ML pipelines. 6) Apply least privilege on the SPARQL backend write account, and use a WAF rule as a stopgap for obvious injection patterns.

What does CISA's SSVC say?

DecisionTrack*
Exploitationpoc
AutomatableNo
Technical Impactpartial

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act
Art. 10-Data and data governance
ISO 42001
A.7.4-Quality of data for AI systemsA.7.5-Data provenance
NIST AI RMF
MAP 4.1-Risks related to third-party data and software are mapped
OWASP LLM Top 10
LLM04-Data and Model Poisoning

Frequently Asked Questions

What is CVE-2026-73975?

djehuty, the research data repository from 4TU.ResearchData, has a SPARQL injection flaw (CWE-943) in a state-modifying DELETE/INSERT query: any authenticated depositor can craft a session name that writes or deletes arbitrary triples across the RDF store shared by all accounts. The result is an integrity compromise of the whole repository's metadata, not only the attacker's own records, and the barrier to entry is low because deployments often allow self-registration through ORCID/SAML. The exploitation signals are modest: EPSS is 0.003 (79th percentile), it is not in CISA KEV, and there is no public exploit or Nuclei template. Even so, this matters if the repository feeds ML datasets, because tampered metadata erodes trust in dataset provenance and labels. Upgrade to djehuty 26.3.2 or later, review the logs for unusual session names and unexpected triple changes, and restore from backup if you suspect manipulation.

Is CVE-2026-73975 actively exploited?

Proof-of-concept exploit code is publicly available for CVE-2026-73975, increasing the risk of exploitation.

How to fix CVE-2026-73975?

1) Upgrade djehuty to version 26.3.2 or later (GHSA-9864-23x8-762h). 2) If you cannot patch right away, restrict or temporarily disable self-registration via ORCID/SAML and review the existing depositor accounts. 3) Review application and SPARQL endpoint logs for session names containing SPARQL syntax characters (quotes, braces, DELETE/INSERT/WHERE keywords). 4) Verify the integrity of the RDF store against a recent backup and restore the affected triples if you find unauthorized changes. 5) Check the checksums and provenance of datasets taken from the repository for ML pipelines. 6) Apply least privilege on the SPARQL backend write account, and use a WAF rule as a stopgap for obvious injection patterns.

What systems are affected by CVE-2026-73975?

This vulnerability affects the following AI/ML architecture patterns: Research data repositories, Dataset management and provenance, RDF/SPARQL triple stores, ML training data pipelines.

What is the CVSS score for CVE-2026-73975?

No CVSS score has been assigned yet.

What is the AI security impact?

Affected AI Architectures

Research data repositoriesDataset management and provenanceRDF/SPARQL triple storesML training data pipelines

MITRE ATLAS Techniques

AML.T0010.002Data
AML.T0012Valid Accounts
AML.T0049Exploit Public-Facing Application
AML.T0059Erode Dataset Integrity

Compliance Controls Affected

EU AI Act: Art. 10
ISO 42001: A.7.4, A.7.5
NIST AI RMF: MAP 4.1
OWASP LLM Top 10: LLM04

What are the technical details?

Original Advisory

djehuty is a research data repository system developed by 4TU.ResearchData. Prior to version 26.3.2, an authenticated depositor can inject arbitrary SPARQL into a state-modifying (DELETE/INSERT) query by supplying a crafted session name, letting them write (and delete) arbitrary triples anywhere in the RDF store. Because the RDF store is shared across all accounts and datasets, this is an integrity compromise of the whole repository's metadata, not just the attacker's own records. Having a logged-in account is a precondition. djehuty allows self-registration via ORCID/SAML, so this is a low barrier in typical deployments. This issue has been patched in version 26.3.2.

Exploitation Scenario

An adversary registers an account through the institution's ORCID/SAML login and creates a session with a crafted name that closes the original SPARQL clause and appends extra DELETE/INSERT statements. When the application runs the state-changing query, the injected statements modify triples belonging to other users: they swap the file references and checksums of a popular dataset, change its license or authorship, or delete the metadata of competing datasets. ML teams that download this dataset to train or benchmark models then ingest altered data without noticing, and the repository loses credibility as a trusted source.

Weaknesses (CWE)

CWE-943 — Improper Neutralization of Special Elements in Data Query Logic: The product generates a query intended to access or manipulate data in a data store such as a database, but it does not neutralize or incorrectly neutralizes special elements that can modify the intended logic of the query.

Source: MITRE CWE corpus.

Timeline

Published
October 1, 2026
Last Modified
October 6, 2026
First Seen
October 1, 2026

Related Vulnerabilities