The AGI Detector is an open-source project designed to be run locally by individual users. Security is a shared responsibility between the maintainers (who provide secure code) and users (who protect their API keys and environment).

1. SSRF Protection: URL validation prevents crawling of localhost, private IPs, and cloud metadata endpoints
2. Safe JSON Parsing: All JSON parsing includes error handling to prevent application crashes
3. Input Validation: API endpoints validate all query parameters and request bodies using Zod schemas
4. Request Limits:
   • Maximum request body size: 1MB
   • Maximum response size: 10MB
   • Request timeouts: 30 seconds
   • Redirect limits: 3 maximum
5. Security Headers:
   • X-Frame-Options: DENY
   • X-Content-Type-Options: nosniff
   • Referrer-Policy: strict-origin-when-cross-origin
   • X-XSS-Protection enabled
6. Browser Security: Playwright launches without dangerous flags (--disable-web-security removed)
7. Dependency Management: Regular dependency updates to patch known vulnerabilities

• Uses Prisma ORM with parameterized queries (SQL injection safe)
• No raw SQL queries with user input
• Connection encryption enforced via SSL/TLS

NEVER commit .env files to version control!

Your .env file contains sensitive credentials:

• OpenAI API key
• Database connection strings
• Brave API key
• Firecrawl API key

Best Practices:

1. Keep .env files local only
2. Use different API keys for development and production
3. Rotate API keys regularly
4. Use read-only database credentials when possible
5. Monitor API usage for unusual activity

The crawler is designed to access public AI research sources only. However:

DO:

• ✅ Use the default trusted sources (OpenAI, DeepMind, Anthropic, etc.)
• ✅ Crawl publicly accessible research blogs and papers
• ✅ Respect rate limits and robots.txt

DON'T:

• ❌ Modify the code to crawl localhost or internal network addresses
• ❌ Crawl private or authenticated endpoints
• ❌ Bypass the URL validator
• ❌ Use the crawler for unauthorized data collection

Run these commands regularly to get security patches:

npm update
npm audit fix

Check for vulnerabilities:

npm audit

If you discover a security vulnerability, please report it responsibly:

DO NOT create a public GitHub issue for security vulnerabilities.

Instead, please:

1. Email the maintainer at: [security contact - add your email]
2. Include:
   • Description of the vulnerability
   • Steps to reproduce
   • Potential impact
   • Suggested fix (if any)

We will:

• Acknowledge receipt within 48 hours
• Provide a fix timeline within 7 days
• Credit you in the fix (unless you prefer to remain anonymous)
• Notify users of the security update

For non-critical security improvements or questions:

• Open a GitHub issue with the label security
• Start a GitHub Discussion in the Security category

Security updates are released as:

• Patch versions (2.0.x) for minor security fixes
• Minor versions (2.x.0) for significant security improvements
• Major versions (x.0.0) for breaking security changes

Subscribe to GitHub releases to get notified of security updates.

This tool is designed for local use only. Do not:

• Deploy API endpoints publicly without authentication
• Expose the database to the internet
• Share your instance with untrusted users

If you need multi-user access, implement:

• User authentication (JWT, OAuth, etc.)
• API key authentication for endpoints
• Rate limiting per user
• Role-based access control

This application relies on third-party services:

• OpenAI API: Data sent to OpenAI for analysis
• Firecrawl: URLs sent to Firecrawl for crawling
• Brave Search: Search queries sent to Brave

Privacy Implications:

• Article content is sent to OpenAI for AGI analysis
• Crawled URLs are visible to Firecrawl
• Search terms are visible to Brave

Recommendations:

• Review OpenAI's data usage policy
• Use API keys with appropriate usage limits
• Consider self-hosted alternatives for sensitive use cases

Before submitting a PR, ensure:

• No hardcoded secrets or API keys
• All user inputs are validated
• SQL queries use parameterized statements
• External URLs are validated before fetching
• Error messages don't leak sensitive information
• Dependencies are up to date (npm audit passes)
• New API endpoints have input validation
• Tests cover security-sensitive code

October 2025 Security Hardening:

• Added SSRF protection with URL validator
• Implemented safe JSON parsing across all API routes
• Added input validation using Zod schemas
• Fixed browser security flags in Playwright
• Updated vulnerable dependencies (axios, babel)
• Added security headers (X-Frame-Options, etc.)
• Implemented request size limits
• Added comprehensive security documentation

• OWASP Top 10
• Next.js Security Best Practices
• Prisma Security Guidelines
• OpenAI API Security

This security policy is part of the AGI Detector project and is licensed under the same MIT License.

Last Updated: October 26, 2025 Policy Version: 1.0