Security Practices

Security isn't a feature at AgentAIShield — it's the foundation. We're in the business of protecting AI agents, which means we hold ourselves to a higher standard than most. This page describes the technical and organizational controls we apply to protect your data and our infrastructure.

1. Infrastructure Security

AgentAIShield is hosted on AWS (primary) and Railway (ancillary services), leveraging enterprise-grade cloud infrastructure with industry-leading physical and network security. All compute is isolated — each service runs in its own container with no shared process space between customers.

• Isolated compute — Services run in dedicated containers; no co-tenancy at the process level
• Encrypted storage — All persistent volumes are encrypted at rest using AES-256
• Private networking — Inter-service communication occurs on isolated VPCs with no public exposure
• DDoS protection — AWS Shield Standard + CloudFront provide volumetric attack mitigation
• Firewall rules — Strict ingress/egress policies enforced at the network layer; only required ports exposed
• Immutable infrastructure — No SSH in production; all changes go through CI/CD pipelines with audit trails

2. Encryption

2.1 In Transit

All data in transit is encrypted using TLS 1.3. We reject connections on TLS 1.0 and 1.1. Our API endpoints enforce HTTPS with HSTS headers and are rated A+ on SSL Labs. Certificate management is automated via Let's Encrypt / AWS ACM with automatic renewal.

2.2 At Rest

All stored data — databases, object storage, and backups — is encrypted using AES-256. Encryption keys are managed in AWS KMS with automatic annual rotation. Database credentials are stored in AWS Secrets Manager, never in environment variables or source code.

2.3 Backups

Automated encrypted backups run daily. Backups are stored in a geographically separate AWS region. Backup encryption keys are distinct from primary data keys. Restore procedures are tested quarterly.

3. Access Control

3.1 Role-Based Access

Internal access to production systems follows strict role-based access control (RBAC). Engineers have access only to the services required for their role. Access is reviewed quarterly and revoked immediately upon offboarding.

3.2 API Key Scoping

Customer API keys are scoped by default to read-only or write-only operations. Enterprise customers can define custom permission scopes per key. All key usage is logged and auditable from the dashboard. Keys are hashed before storage — we never store plaintext API keys.

3.3 MFA Support

All internal staff accounts require multi-factor authentication (MFA) enforced at the IdP level. Customer accounts support TOTP-based MFA (available on all plans) and SSO/SAML (Business and Enterprise). We strongly recommend enabling MFA on all customer accounts.

3.4 Privileged Access

Production database access requires time-limited credentials issued via a PAM system, with every session logged. No standing privileged access exists — engineers request access, it is granted for the task duration, then automatically revoked.

4. Data Isolation

AgentAIShield is a multi-tenant platform with strict organization-level data isolation. Each organization's data — API logs, Trust Scores, configurations, and billing records — is logically isolated by organization ID, enforced at both the application and database query layers.

• No cross-tenant data access — It is architecturally impossible for one organization to query another's data through our APIs
• Row-level security — PostgreSQL row-level security policies enforce org isolation at the database layer as a second line of defense
• Separate S3 prefixes — Object storage is partitioned by organization ID with distinct IAM policies per prefix
• API key namespacing — All keys are validated against the requesting organization before any data is returned

5. SOC 2 Compliance

We are actively on our SOC 2 Type II compliance journey. Our controls are designed around the AICPA Trust Service Criteria: Security, Availability, Processing Integrity, Confidentiality, and Privacy. Milestones:

• Controls implemented — All core SOC 2 controls are in place, documented, and monitored
• Continuous monitoring — Automated compliance monitoring via Drata tracks control health in real time
• Annual audits planned — We plan to complete our first formal SOC 2 Type II audit within 12 months of launch
• Enterprise customers — May request our current security documentation and in-progress audit evidence under NDA

6. Penetration Testing

We take a proactive approach to finding vulnerabilities before adversaries do:

• Third-party penetration tests — We engage independent security firms to conduct full-scope penetration tests annually, covering API endpoints, authentication flows, and infrastructure
• Internal security reviews — Every major feature undergoes a security design review before deployment
• Automated SAST/DAST — Static analysis (via CodeQL) and dynamic scanning run on every pull request
• Dependency scanning — Dependabot and Snyk monitor all dependencies for known CVEs and auto-open PRs for patches
• Bug bounty program — A formal bug bounty program with cash rewards for qualifying vulnerability reports is planned for Q3 2026

7. Incident Response

We maintain a documented Incident Response Plan (IRP) that is tested semi-annually via tabletop exercises. Our response commitments:

• 24-hour response SLA — All confirmed security incidents receive a dedicated incident commander within 24 hours of detection
• Severity tiers — P0 (active breach), P1 (data exposure risk), P2 (degraded security control), P3 (informational) — each with defined escalation paths and timelines
• Customer notification — Affected customers are notified within 72 hours of a confirmed data exposure event, per GDPR Article 33 and applicable state laws
• Post-mortem for every incident — Every P0/P1 incident results in a written post-mortem with root cause analysis and remediation items, shared with affected customers upon request
• Status page — Real-time incident status is published at agentaishield.com/status.html

8. Vulnerability Disclosure

We believe in coordinated disclosure and welcome reports from the security community. Our responsible disclosure policy:

• Report to: [email protected]
• PGP key: Available upon request for encrypted submissions
• Acknowledgment: We acknowledge receipt within 2 business days
• Assessment: We provide an initial assessment and severity rating within 5 business days
• Safe harbor: Good-faith security research conducted according to this policy will not result in legal action
• No public disclosure without coordinating a remediation timeline with us first (standard 90-day window)
• Hall of Fame: Researchers who report valid vulnerabilities are credited in our security acknowledgements (with permission)

9. Data Retention

We don't hold your data longer than necessary:

• Request logs & monitoring data — Configurable retention per plan: 30 days (Free), 30 days (Starter), 90 days (Business), 1 year (Enterprise)
• Data export — Export all your monitoring data in JSON or CSV format at any time from Settings → Data Export
• Right to deletion — Delete your account and all associated data at any time. Deletion completes within 30 days. Backups are purged within 90 days.
• Automated purging — A scheduled job enforces retention policies nightly; data is not manually archived beyond configured limits

10. Employee Security

Our people are our first line of defense — and a potential attack surface. We address both:

• Background checks — All full-time employees and contractors with production access undergo background screening prior to being granted access
• Security training — Annual security awareness training is mandatory for all staff, covering phishing, social engineering, and secure coding practices
• Least-privilege access — Every team member receives only the minimum access required to perform their role, reviewed quarterly
• Secure endpoints — Company devices are enrolled in MDM, encrypted, and require screen lock. Personal devices may not access production systems.
• Offboarding — Access revocation is automated and completed within 1 hour of employment termination
• Confidentiality agreements — All staff sign NDAs and data handling agreements before day one