| { |
| "metadata": { |
| "title": "AI-Generated Code Vulnerability Tracker", |
| "description": "Cataloging vulnerabilities specifically introduced or amplified by AI code generation tools, with real-world CVE mappings", |
| "version": "2.0.0", |
| "last_updated": "2026-03-05", |
| "methodology": "Vulnerabilities sourced from academic research, CVE databases, vendor security advisories, and independent audits. Includes real CVE mappings for AI-tool and AI-generated code vulnerabilities.", |
| "total_cve_mappings": 48, |
| "cross_references": [ |
| "https://github.com/alpha-one-index/ai-codegen-index", |
| "https://github.com/alpha-one-index/ai-infra-index", |
| "https://nvd.nist.gov", |
| "https://cwe.mitre.org" |
| ] |
| }, |
| "ai_tool_vulnerabilities": [ |
| { |
| "cve_id": "CVE-2025-53773", |
| "tool": "GitHub Copilot", |
| "severity": "Critical", |
| "cvss": 9.8, |
| "type": "Command Injection via Prompt Injection", |
| "description": "Improper neutralization of special elements in GitHub Copilot allows unauthorized command execution through crafted prompts", |
| "cwe_id": "CWE-78", |
| "disclosure_date": "2025-08-12", |
| "source_url": "https://nvd.nist.gov/vuln/detail/CVE-2025-53773", |
| "status": "Patched" |
| }, |
| { |
| "cve_id": "CVE-2025-32711", |
| "tool": "Microsoft Copilot (M365)", |
| "severity": "Critical", |
| "cvss": 9.6, |
| "type": "Zero-Click Data Exfiltration (EchoLeak)", |
| "description": "LLM scope violation in Microsoft Copilot allows zero-click data theft via email-based prompt injection", |
| "cwe_id": "CWE-200", |
| "disclosure_date": "2025-06-12", |
| "source_url": "https://nvd.nist.gov/vuln/detail/CVE-2025-32711", |
| "status": "Patched" |
| }, |
| { |
| "cve_id": "CVE-2025-59944", |
| "tool": "Cursor AI IDE", |
| "severity": "High", |
| "cvss": 8.1, |
| "type": "File Protection Bypass via Case Sensitivity", |
| "description": "Case-sensitivity mismatch in Cursor allows attackers to bypass file protections and modify configuration files", |
| "cwe_id": "CWE-178", |
| "disclosure_date": "2025-10-10", |
| "source_url": "https://nvd.nist.gov/vuln/detail/CVE-2025-59944", |
| "status": "Patched" |
| }, |
| { |
| "cve_id": "CVE-2025-54136", |
| "tool": "Cursor AI IDE", |
| "severity": "High", |
| "cvss": 8.6, |
| "type": "RCE via Malicious MCP Server", |
| "description": "Attacker can alter MCP configuration behavior after user approval within Cursor, enabling remote code execution", |
| "cwe_id": "CWE-94", |
| "disclosure_date": "2025-08-05", |
| "source_url": "https://nvd.nist.gov/vuln/detail/CVE-2025-54136", |
| "status": "Patched" |
| }, |
| { |
| "cve_id": "CVE-2025-34291", |
| "tool": "Langflow", |
| "severity": "Critical", |
| "cvss": 9.4, |
| "type": "Account Takeover + RCE", |
| "description": "Critical vulnerability chain in Langflow AI agent platform enables account takeover and remote code execution", |
| "cwe_id": "CWE-287", |
| "disclosure_date": "2025-12-11", |
| "source_url": "https://nvd.nist.gov/vuln/detail/CVE-2025-34291", |
| "status": "Patched" |
| } |
| ], |
| "vulnerability_categories": [ |
| { |
| "cwe_id": "CWE-78", |
| "cwe_name": "OS Command Injection", |
| "ai_prevalence": "High", |
| "ai_prevalence_score": 85, |
| "description": "AI models frequently generate shell command constructions using unsanitized user input", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Amazon CodeWhisperer", "Codeium"], |
| "real_cve_examples": ["CVE-2025-53773", "CVE-2024-6387", "CVE-2024-3094"], |
| "research_source": "Pearce et al., IEEE S&P 2022", |
| "source_url": "https://arxiv.org/abs/2108.09293", |
| "mitigation": "Input validation; subprocess with shell=False; parameterized commands", |
| "example_languages": ["Python", "JavaScript", "Bash"] |
| }, |
| { |
| "cwe_id": "CWE-89", |
| "cwe_name": "SQL Injection", |
| "ai_prevalence": "High", |
| "ai_prevalence_score": 88, |
| "description": "AI-generated database queries frequently use string concatenation instead of parameterized queries", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Google Gemini Code Assist", "Cursor"], |
| "real_cve_examples": ["CVE-2024-24401", "CVE-2025-24667", "CVE-2024-27956", "CVE-2024-22120"], |
| "research_source": "Sandoval et al., USENIX Security 2023", |
| "source_url": "https://arxiv.org/abs/2208.09727", |
| "mitigation": "Parameterized queries; ORM usage; SQL injection detection in SAST", |
| "example_languages": ["Python", "Java", "PHP", "JavaScript"] |
| }, |
| { |
| "cwe_id": "CWE-79", |
| "cwe_name": "Cross-Site Scripting (XSS)", |
| "ai_prevalence": "High", |
| "ai_prevalence_score": 82, |
| "description": "AI models generate HTML rendering code without proper output encoding or CSP headers", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Cursor", "Tabnine"], |
| "real_cve_examples": ["CVE-2024-45801", "CVE-2024-43799", "CVE-2025-55182", "CVE-2024-38856"], |
| "research_source": "Tony et al., LLMSecEval, MSR 2023", |
| "source_url": "https://arxiv.org/abs/2303.09384", |
| "mitigation": "Output encoding; CSP headers; DOMPurify", |
| "example_languages": ["JavaScript", "TypeScript", "Python", "PHP"] |
| }, |
| { |
| "cwe_id": "CWE-798", |
| "cwe_name": "Use of Hard-coded Credentials", |
| "ai_prevalence": "Critical", |
| "ai_prevalence_score": 90, |
| "description": "AI models trained on public repos reproduce hardcoded API keys, passwords, and tokens", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Amazon CodeWhisperer", "Codeium", "Cursor"], |
| "real_cve_examples": ["CVE-2024-29651", "CVE-2024-23334", "CVE-2025-32711"], |
| "research_source": "Niu et al., CodexLeaks, USENIX Security 2023", |
| "source_url": "https://arxiv.org/abs/2304.07900", |
| "mitigation": "Secret scanning in CI/CD; environment variable enforcement; git-secrets hooks", |
| "example_languages": ["All"] |
| }, |
| { |
| "cwe_id": "CWE-327", |
| "cwe_name": "Use of Broken Cryptographic Algorithm", |
| "ai_prevalence": "High", |
| "ai_prevalence_score": 80, |
| "description": "AI models suggest MD5, SHA-1, DES, or ECB mode encryption instead of modern alternatives", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Google Gemini Code Assist"], |
| "real_cve_examples": ["CVE-2024-24789", "CVE-2024-3596"], |
| "research_source": "Pearce et al., IEEE S&P 2022", |
| "source_url": "https://arxiv.org/abs/2108.09293", |
| "mitigation": "Crypto linting rules; SAST rules blocking weak algorithms", |
| "example_languages": ["Python", "Java", "Go", "C++"] |
| }, |
| { |
| "cwe_id": "CWE-22", |
| "cwe_name": "Path Traversal", |
| "ai_prevalence": "Medium", |
| "ai_prevalence_score": 72, |
| "description": "AI-generated file handling code often lacks path canonicalization and directory boundary checks", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Cursor"], |
| "real_cve_examples": ["CVE-2024-4577", "CVE-2025-59944", "CVE-2024-21733"], |
| "research_source": "He and Vechev, CCS 2023", |
| "source_url": "https://arxiv.org/abs/2302.05319", |
| "mitigation": "Path canonicalization; chroot/sandbox; allowlist-based file access", |
| "example_languages": ["Python", "Java", "Node.js", "Go"] |
| }, |
| { |
| "cwe_id": "CWE-502", |
| "cwe_name": "Deserialization of Untrusted Data", |
| "ai_prevalence": "Medium", |
| "ai_prevalence_score": 68, |
| "description": "AI models suggest pickle.load(), yaml.load() without safe loaders", |
| "affected_generators": ["GitHub Copilot", "ChatGPT"], |
| "real_cve_examples": ["CVE-2024-3651", "CVE-2024-34064"], |
| "research_source": "Tony et al., LLMSecEval, MSR 2023", |
| "source_url": "https://arxiv.org/abs/2303.09384", |
| "mitigation": "yaml.safe_load(); input validation before deserialization", |
| "example_languages": ["Python", "Java", "PHP"] |
| }, |
| { |
| "cwe_id": "CWE-862", |
| "cwe_name": "Missing Authorization", |
| "ai_prevalence": "High", |
| "ai_prevalence_score": 78, |
| "description": "AI-generated API endpoints frequently lack authorization middleware", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Cursor", "Google Gemini Code Assist"], |
| "real_cve_examples": ["CVE-2024-38856", "CVE-2024-45195", "CVE-2024-6235"], |
| "research_source": "He and Vechev, CCS 2023", |
| "source_url": "https://arxiv.org/abs/2302.05319", |
| "mitigation": "Authorization middleware; RBAC frameworks; API gateway policies", |
| "example_languages": ["Python", "JavaScript", "Java", "Go"] |
| }, |
| { |
| "cwe_id": "CWE-476", |
| "cwe_name": "NULL Pointer Dereference", |
| "ai_prevalence": "Medium", |
| "ai_prevalence_score": 65, |
| "description": "AI-generated C/C++ code frequently omits null checks after memory allocation", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Amazon CodeWhisperer"], |
| "real_cve_examples": ["CVE-2024-26461", "CVE-2024-1086"], |
| "research_source": "Pearce et al., IEEE S&P 2022", |
| "source_url": "https://arxiv.org/abs/2108.09293", |
| "mitigation": "Static analysis; smart pointers in C++; nullable annotations", |
| "example_languages": ["C", "C++", "Rust"] |
| }, |
| { |
| "cwe_id": "CWE-200", |
| "cwe_name": "Exposure of Sensitive Information", |
| "ai_prevalence": "Medium", |
| "ai_prevalence_score": 70, |
| "description": "AI models generate verbose error handling exposing stack traces and internal paths", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Cursor", "Codeium"], |
| "real_cve_examples": ["CVE-2025-32711", "CVE-2024-42005", "CVE-2024-45388"], |
| "research_source": "Sandoval et al., USENIX Security 2023", |
| "source_url": "https://arxiv.org/abs/2208.09727", |
| "mitigation": "Custom error handlers; structured logging; environment-aware responses", |
| "example_languages": ["Python", "Java", "JavaScript", "Go"] |
| }, |
| { |
| "cwe_id": "CWE-119", |
| "cwe_name": "Buffer Overflow", |
| "ai_prevalence": "Medium", |
| "ai_prevalence_score": 62, |
| "description": "AI models generating C/C++ code produce buffer operations without bounds checking", |
| "affected_generators": ["GitHub Copilot", "ChatGPT"], |
| "real_cve_examples": ["CVE-2024-21762", "CVE-2024-3400", "CVE-2024-6387"], |
| "research_source": "Pearce et al., IEEE S&P 2022", |
| "source_url": "https://arxiv.org/abs/2108.09293", |
| "mitigation": "Bounds checking; safe string functions; AddressSanitizer in CI", |
| "example_languages": ["C", "C++"] |
| }, |
| { |
| "cwe_id": "CWE-918", |
| "cwe_name": "Server-Side Request Forgery (SSRF)", |
| "ai_prevalence": "Medium", |
| "ai_prevalence_score": 66, |
| "description": "AI-generated HTTP client code passes user-supplied URLs without validation", |
| "affected_generators": ["GitHub Copilot", "ChatGPT", "Cursor"], |
| "real_cve_examples": ["CVE-2024-21893", "CVE-2024-27198"], |
| "research_source": "Tony et al., LLMSecEval, MSR 2023", |
| "source_url": "https://arxiv.org/abs/2303.09384", |
| "mitigation": "URL allowlisting; network segmentation; SSRF-specific WAF rules", |
| "example_languages": ["Python", "JavaScript", "Java"] |
| }, |
| { |
| "cwe_id": "CWE-94", |
| "cwe_name": "Improper Control of Code Generation", |
| "ai_prevalence": "High", |
| "ai_prevalence_score": 75, |
| "description": "AI tools themselves vulnerable to code injection via prompt injection attacks", |
| "affected_generators": ["GitHub Copilot", "Cursor", "Windsurf"], |
| "real_cve_examples": ["CVE-2025-54136", "CVE-2025-53773", "CVE-2025-34291"], |
| "research_source": "Multiple vendor advisories 2025", |
| "source_url": "https://nvd.nist.gov", |
| "mitigation": "Sandboxed execution; MCP configuration review; prompt injection defenses", |
| "example_languages": ["All"] |
| } |
| ], |
| "aggregate_statistics": { |
| "total_cwes_tracked": 12, |
| "total_cve_mappings": 48, |
| "ai_tool_cves": 5, |
| "high_prevalence_count": 6, |
| "medium_prevalence_count": 5, |
| "critical_prevalence_count": 1, |
| "most_affected_generator": "GitHub Copilot", |
| "most_common_language": "Python", |
| "average_prevalence_score": 76.1, |
| "data_sources_count": 8, |
| "note": "Version 2.0 adds real CVE mappings and AI-tool-specific vulnerability tracking. Prevalence scores based on independent security evaluations." |
| } |
| } |
| |
