VAITP Dataset

Dataset Statistics

Browse the dataset

  • 2,412 vulnerabilities
  • 2,412 vulnerable / patched code pairs
  • 930 exploit payloads
  • 10 categories
  • newest CVE

Showing 51โ€“75 of 143 vulnerabilities matching your filters

Download these results: CSV file JSON file

VAITP dataset entries, page 3 of 6, sorted by recently added
CVE Summary Category CWE Scope CVSS Published
CVE-2026-45758 Malicious guardrails-ai version 0.10.1 on PyPI could expose credentials. Design Defects Vulnerable and Outdated Components CWE-506 Remote 9.6 Critical
CVE-2026-43986 Tautulli < 2.17.1 has an unauthenticated SSRF via its image proxy. Design Defects Server-Side Request Forgery (SSRF) CWE-918 Remote 9.9 Critical
CVE-2026-5241 Hugging Face LightGlue ignores `trust_remote_code`, allowing RCE on load. Design Defects Insecure Parsing or Deserialization CWE-829 Remote 9.6 Critical
CVE-2026-33234 SSRF in AutoGPT's SendEmailBlock allows internal network port scanning. Design Defects Server-Side Request Forgery (SSRF) CWE-918 Remote 5.0 Medium
CVE-2023-54345 ERPNext sandbox escape allows System Managers to execute arbitrary code. Design Defects Command Injection CWE-94 Remote 8.7 High
CVE-2026-44827 RCE in Diffusers via malicious `None.py` file when loading Hub pipelines. Design Defects Time-of-Check to Time-of-Use CWE-94 Remote 8.8 High
CVE-2026-45227 Heym sandbox escape via Python introspection allows arbitrary command execution. Design Defects Command Injection CWE-693 Remote 8.7 High
CVE-2026-31231 RCE in Cognee <=v0.4.0's notebook API due to unsandboxed code execution. Design Defects Command Injection CWE-94 Remote 9.8 Critical
CVE-2026-6357 pip's post-install update check may execute code from a malicious package. Design Defects Time-of-Check to Time-of-Use CWE-829 Local 5.3 Medium
CVE-2026-40602 Unsandboxed Jinja2 template in hass-cli < 1.0.0 allows code execution. Design Defects Command Injection CWE-94 Local 5.6 Medium
CVE-2026-39419 MaxKB sandbox bypass allows authenticated users to spoof tool results. Design Defects Poorly Designed Access Controls CWE-74 Remote 3.1 Low
CVE-2026-39420 Authenticated users can escape MaxKB's sandbox for RCE using 'env -i'. Design Defects Poorly Designed Access Controls CWE-78 Remote 7.4 High
CVE-2026-39888 PraisonAI sandbox escape via frame traversal allows remote code execution. Design Defects Command Injection CWE-657 Remote 9.9 Critical
CVE-2026-2275 CrewAI CodeInterpreter's unsafe fallback to SandboxPython enables RCE. Design Defects Inadequate Error Handling CWE-749 Remote 9.6 Critical
CVE-2026-28505 Tautulli sandbox escape in notification templates using lambda expressions. Design Defects Command Injection CWE-94 Remote 7.5 High
CVE-2026-33873 Langflow Agentic Assistant allows RCE via LLM-generated Python code. Design Defects Command Injection CWE-94 Remote 9.3 Critical
CVE-2026-27448 pyOpenSSL: Unhandled exception in servername callback leads to security bypass. Design Defects Inadequate Error Handling CWE-636 Remote 1.7 Low
CVE-2026-32128 FastGPT sandbox file write guardrail bypass via stdout remapping. Design Defects Poorly Designed Access Controls CWE-184 Local 6.3 Medium
CVE-2026-2297 CPython's import of legacy .pyc files bypasses sys.audit events. Design Defects Poorly Designed Access Controls CWE-668 Local 5.7 Medium
CVE-2026-27952 Agenta sandbox escape via whitelisted numpy allows code execution. Design Defects Command Injection CWE-94 Remote 9.9 Critical
CVE-2026-27494 n8n Python Code node sandbox escape allows for RCE and file exfiltration. Design Defects Poorly Designed Access Controls CWE-497 Remote 7.1 High
CVE-2026-25115 n8n Python Code node sandbox escape allows authenticated code execution. Design Defects Privilege Escalation CWE-693 Remote 9.4 Critical
CVE-2026-24002 Insecure Grist pyodide sandbox allows for arbitrary code execution. Design Defects Security Misconfigurations CWE-74 Remote 9.6 Critical
CVE-2026-22807 Untrusted code execution in vLLM via malicious Hugging Face model repos. Design Defects Remote File Inclusion (RFI) CWE-94 Remote 9.8 Critical
CVE-2025-14026 Forcepoint One DLP Client ctypes restriction bypass allows code execution. Design Defects Poorly Designed Access Controls CWE-913 Local 7.8 High

Download the dataset

Every visible entry with all of its fields: classification, descriptions, the vulnerable and patched code pair, and the exploit payload where there is one. Current files: 2,411 entries, built .

SHA-256 checksums
00100ae17adddda8f786a502580a64470cdaa8e95d6d33eded83ffdbdfc63386  vaitp-dataset-2026-11-21.csv
77624d5fe7cf1f5e7dde13d81a8186fd8fc520a506b7dedf831e3ffc6a232722  vaitp-dataset-2026-11-21.json

Licence and citation

The VAITP dataset is published under the Creative Commons Attribution 4.0 International (CC BY 4.0) licence: you may share and adapt it for any purpose, with attribution.

Bogaerts, F., Ivaki, N., & Fonseca, J. (2026). VAITP Python Vulnerability Dataset [Data set]. https://netpack.pt/vaitp/dataset/

@misc{vaitp_dataset_2026,
  title        = {{VAITP} Python Vulnerability Dataset},
  author       = {Bogaerts, Fr\'ed\'eric and Ivaki, Naghmeh and Fonseca, Jos\'e},
  year         = {2026},
  howpublished = {\url{https://netpack.pt/vaitp/dataset/}},
  note         = {Licensed under CC BY 4.0}
}
What each field means
Fields in the downloads
FieldMeaning
idInternal VAITP record number.
cveCVE identifier.
cve_urlLink to the CVE record.
cweCWE weakness as stored, e.g. "CWE-79" or "CWE-94 (inferred)" when assigned by the labelling model rather than NVD.
cwe_numberNumeric CWE identifier, when there is one.
cwe_urlLink to the CWE definition at MITRE.
cvss_scoreCVSS base score from NVD (newest version available). Empty when NVD has not scored it.
cvss_ratingQualitative rating of the score: low, medium, high or critical.
publishedCVE publication date (YYYY-MM-DD).
owaspOWASP Top 10 category.
odcOrthogonal Defect Classification defect type.
cdcCode defect classification (VAITP taxonomy).
categoryVulnerability category (VAITP taxonomy).
subcategoryVulnerability subcategory (VAITP taxonomy).
scopeAccessibility scope: whether the flaw is reachable remotely or only locally.
impactMain impact of a successful exploit.
summaryOne-line description.
descriptionFull CVE description.
solutionHow the vulnerability is fixed or mitigated.
affected_componentLibrary, package or component the change is in.
fixed_by_upgradingWhether upgrading the affected component fixes it.
vulnerable_codePython sample containing the vulnerability.
patched_codeThe same sample with the vulnerability fixed.
payload_neededWhether exploiting the sample needs a crafted input.
payloadProof-of-concept input that exploits the vulnerable sample, when available.
payload_statusavailable, none, or model_refused (the labelling model declined to write one).
referencesFurther references.
permalinkThis entry's page on the VAITP website.
Use the API instead

The same filters work on the read-only JSON API, so you can script against it:

https://netpack.pt/vaitp/wp-json/vaitp/v1/vulnerabilities?category=Cryptographic&min_score=7&per_page=100
https://netpack.pt/vaitp/wp-json/vaitp/v1/vulnerabilities/CVE-2024-1234
https://netpack.pt/vaitp/wp-json/vaitp/v1/stats
https://netpack.pt/vaitp/wp-json/vaitp/v1/export?format=csv&q=flask
Introducing the "VAITP dataset": a specialized repository of Python vulnerabilities and patches, meticulously compiled for the use of the security research community. As Python's prominence grows, understanding and addressing potential security vulnerabilities become crucial. Crafted by and for the cybersecurity community, this dataset offers a valuable resource for researchers, analysts, and developers to analyze and mitigate the security risks associated with Python. Through the comprehensive exploration of vulnerabilities and corresponding patches, the VAITP dataset fosters a safer and more resilient Python ecosystem, encouraging collaborative advancements in programming security.

The supreme art of war is to subdue the enemy without fighting.

Sun Tzu – “The Art of War”

:: Shaping the future through research and ingenuity ::

Before you use the VAITP AI models

Please read and accept the following disclaimer:

By choosing "I understand and accept", you acknowledge that you understand and agree to these terms.