Your inputs never train AI models
Customer inputs are not used to train the underlying AI models. Conversations are retained to operate the service, support safety, and improve reliability — never resold, never used for model training.
security & data practices
built for agencies that answer to the public.
Your evaluators, your counsel, and your records custodian all get a straight answer here. This page is written to be printed and put in the file.
No PII or PHI stored by design
Cass works from role descriptions, not names. When a user pastes identifying details, drafts are written without propagating them, and the after-action record is written for the public record. Personal health information has no place in the product and none is stored.
A human approves every word
Nothing auto-fires. Every statement, release, alert, and social post requires the credentialed PIO's explicit approval before it leaves the agency. Cass is disclosed as AI everywhere she appears.
Encryption in transit, least-privilege access
Connections to the site and between our systems use TLS. Internal access follows the principle of least privilege with audit logging; administrative surfaces sit behind additional access controls beyond the product login.
Account security
Sign-in uses one-time email codes or Google sign-in — no passwords for us to lose. Sessions are hardened server-side and can be revoked.
US-based infrastructure
The product runs on Cloudflare's platform with data stored in the United States. Automated backups run on a regular schedule.
Built for the public record
Released statements are treated as public records — retained, timestamped, and never silently purged. The activity log is your ICS-214; exports exist because records requests are a reality of this work, not an afterthought.
Published sub-processors
Every third party that touches service data is named publicly, with what it does and why — including our AI provider (Anthropic), transcription (Deepgram), email (Resend), and infrastructure (Cloudflare).
certifications: none claimed. on purpose.
We don't hold SOC 2, ISO 27001, or FedRAMP today, and we won't imply otherwise with badge walls. What we publish is what we actually do — the practices above, the named sub-processors, and contract terms written for public agencies. As we earn formal attestations, they'll be published here.
Security question, concern, or something you believe you've found? cass@heycass.ai reaches a human quickly.