We release patches for security vulnerabilities in the following versions:

We take security bugs seriously. We appreciate your efforts to responsibly disclose your findings, and will make every effort to acknowledge your contributions.

Please do not report security vulnerabilities through public GitHub issues.

Instead, please report them via email to:

security@redasgard.com

When reporting a security vulnerability, please include:

1. Description: A clear description of the vulnerability
2. Steps to Reproduce: Detailed steps to reproduce the issue
3. Impact: Description of the potential impact
4. Environment: OS, Rust version, LLM model, and any other relevant details
5. Proof of Concept: If possible, include a minimal code example that demonstrates the issue

• Acknowledgment: We will acknowledge receipt of your report within 48 hours
• Initial Assessment: We will provide an initial assessment within 5 business days
• Regular Updates: We will keep you informed of our progress
• Resolution: We will work with you to resolve the issue and coordinate disclosure

• Initial Response: Within 48 hours
• Status Update: Within 5 business days
• Resolution: Within 30 days (depending on complexity)

When reporting vulnerabilities, please consider:

1. Prompt Injection: Successful bypass of injection detection
2. Jailbreak Attacks: Successful jailbreak attempts
3. Unicode Attacks: Bypass through Unicode manipulation
4. False Positives: Legitimate inputs being blocked
5. Performance: DoS through resource exhaustion
6. Memory Safety: Unsafe memory operations or buffer overflows

Common attack vectors to test:

• Direct Injection: "Ignore previous instructions"
• Jailbreak Techniques: DAN, STAN, developer mode
• Unicode Homoglyphs: Visually similar characters
• Zero-Width Characters: Hidden characters
• RTL Override: Right-to-left text manipulation
• Encoding Attacks: URL encoding, HTML entities
• Social Engineering: Professional-sounding manipulation
• Semantic Cloaking: Disguised attack patterns

1. Always validate inputs before sending to LLMs
2. Use the library correctly according to documentation
3. Implement additional layers of security
4. Keep the library updated to the latest version
5. Monitor for security advisories
6. Test with your specific LLM models

1. Test with malicious inputs regularly
2. Implement defense in depth
3. Use the library correctly according to documentation
4. Consider additional validation for critical applications
5. Monitor security updates
6. Test with various LLM models

• 90+ Attack Patterns: Comprehensive pattern matching
• Multi-Layer Detection: Input sanitization and output validation
• Unicode Protection: Homoglyph and zero-width character detection
• Jailbreak Prevention: DAN, STAN, and other jailbreak techniques
• Memory Safety: Rust's memory safety guarantees
• Configurable Security: Adjustable security levels

• Input Sanitization: Sanitize user input before validation
• Output Validation: Validate LLM responses
• Rate Limiting: Implement rate limiting for LLM calls
• Logging: Log security events for monitoring
• Regular Updates: Keep dependencies and the library updated
• Model Testing: Test with your specific LLM models

1. Assessment: We assess the severity and impact
2. Fix Development: We develop a fix in private
3. Testing: We thoroughly test the fix
4. Release: We release the fix with a security advisory
5. Disclosure: We coordinate disclosure with reporters

Security advisories are published on:

• GitHub Security Advisories: https://github.com/redasgard/llm-security/security/advisories
• Crates.io: Security notices in release notes
• Email: Subscribers to security@redasgard.com

We follow responsible disclosure practices:

1. Private Reporting: Report vulnerabilities privately first
2. Coordinated Disclosure: We coordinate disclosure timing
3. Credit: We give credit to security researchers
4. No Legal Action: We won't take legal action against good faith research

• Test Responsibly: Don't test on production systems
• Respect Privacy: Don't access or modify data
• Report Promptly: Report findings as soon as possible
• Follow Guidelines: Follow this security policy

In Scope:

• Prompt injection bypasses
• Jailbreak technique bypasses
• Unicode manipulation attacks
• False positive issues
• Memory safety issues
• Performance DoS attacks

Out of Scope:

• Social engineering attacks
• Physical security issues
• Issues in dependencies (report to their maintainers)
• Issues in applications using this library
• Issues in LLM models themselves

For security-related questions or to report vulnerabilities:

• Email: security@redasgard.com
• PGP Key: Available upon request
• Response Time: Within 48 hours

We thank the security researchers who help keep our software secure. Security researchers who follow responsible disclosure practices will be acknowledged in our security advisories.

By reporting a security vulnerability, you agree to:

1. Not disclose the vulnerability publicly until we've had a chance to address it
2. Not access or modify data that doesn't belong to you
3. Not disrupt our services or systems
4. Act in good faith to avoid privacy violations, destruction of data, and interruption or degradation of our services

Thank you for helping keep LLM Security and our users safe! 🤖🛡️