Clinical Capture

Security Overview
← Home

Security Overview

Clinical Capture is built around a simple principle that minimizes risk: it does not store the clinical text it analyzes. The summary below is intended for security and IT reviewers; a more detailed packet (data-flow diagram, control inventory, and questionnaire responses) is available on request via our contact form.

Data handling

  • Operative note text is processed transiently and is never written to persistent storage.
  • Only de-identified notes are intended to be entered; users acknowledge this at registration.
  • We store account data and anonymous usage metadata (no note content).
  • Note text submitted via our AI provider's API is not used to train models.

Encryption

  • All traffic is encrypted in transit with TLS (HTTPS enforced; HSTS enabled).
  • Data at rest (account and metadata) is encrypted by our database provider.
  • Passwords are hashed with bcrypt; secrets are stored as encrypted environment variables.

Access & authentication

  • Access is invite-gated; each request re-checks authentication and authorization.
  • Optional two-factor authentication (TOTP) with backup codes.
  • Brute-force protection: account lockout after repeated failed sign-ins.
  • Sessions expire 12 hours after sign-in.
  • Sign-in activity is logged for audit.

Application security

  • Content Security Policy, X-Frame-Options (DENY), nosniff, and related headers enforced.
  • Built on a modern, actively maintained framework with managed dependency updates.
  • No operative note content is written to application or error logs.

Subprocessors

  • Vercel — application hosting and compute (SOC 2). Processes note text transiently.
  • Anthropic — AI model provider (SOC 2; BAA available). Processes note text transiently; no training on inputs.
  • Neon — encrypted Postgres for account/metadata only. Does not receive note text.
  • Resend — transactional email only. Does not receive note text.

Intended use & scope

Clinical Capture is administrative billing-and-coding decision support with a required human-verification step. It is not a medical device and is not, in its current phase, intended to process identifiable PHI. See our Privacy Policy and Terms of Use.

Responsible disclosure

Security researchers may report concerns via our contact form or the address in our security.txt.