Security Practices
How We Handle Sensitive Work
Baseline practices for AI red teaming, assessment, advisory, and incident-support engagements.
Least-Privilege Access
Client access is scoped to the engagement and removed when no longer needed.
Evidence Handling
Findings, logs, and screenshots are handled as confidential engagement materials.
Model & Credential Safety
We avoid collecting secrets, model weights, and training data whenever possible, and use agreed secure channels when access is required.
Incident Escalation
Severity, ownership, and communications are defined during onboarding or active incident scoping.