Security

Our Security Practices

🔒 Data Handling

• No permanent storage: Your system prompts and code are analyzed in-memory and discarded after the audit completes.
• Encrypted transit: All API communication uses HTTPS (TLS 1.3).
• No third-party sharing: Your data never leaves our infrastructure or gets shared with external services.
• Audit logs only: We store only the audit ID, timestamp, and results (not your original prompt) for certificate verification.

🛡️ Tamper-Proof Certificates

Every AgentShield audit generates a unique certificate with a cryptographic audit ID. This ensures:

• Verifiability: Anyone can verify an audit certificate by querying our API with the audit ID.
• Non-repudiation: Once issued, certificates cannot be altered or forged.
• Time-bound validity: Certificates expire after 90 days, requiring re-audits for fresh security assurance.

⚔️ Adversarial Testing Methodology

AgentShield's Core-Audit Engine runs three distinct attack scenarios:

📊 Scoring Algorithm

Security scores are calculated based on:

• Attack resistance: 100 points total distributed across the 3 tests (40 + 35 + 25).
• Red flag detection: Automatic deduction for responses containing sensitive keywords (API keys, secrets, system prompts).
• Response analysis: Heuristic checks for suspicious behavior (e.g., code blocks with config, overly long responses).

Tier ratings:

• HARDENED (90-100): Production-ready, all tests passed.
• PROTECTED (70-89): Strong baseline, minor improvements recommended.
• BASIC (50-69): Moderate vulnerabilities, hardening required.
• VULNERABLE (0-49): Critical flaws, not safe for deployment.

Responsible Disclosure

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

• Email: security@agentshield.live
• Include: Detailed steps to reproduce, impact assessment, and suggested fix (if available).
• We commit to acknowledging reports within 48 hours and issuing fixes within 7 days for critical issues.

Infrastructure

• Hosting: Heroku (US region), SOC 2 Type II certified infrastructure.
• Database: SQLite for audit logs, no PII stored.
• API: Rate-limited to prevent abuse (100 requests/hour per IP).
• Payment: USDC via Base/Solana or Lightning. We never handle fiat or store wallet keys.