Security
Last updated: September 24, 2026
This page describes the controls AutoTalent has in place today, as built. We’re a small, early company, so we also list what we don’t have yet.
Infrastructure
- The web app and its API run on Vercel. The MCP server used by the CLI and agents runs on Railway.
- Data is stored in a managed PostgreSQL database on Supabase, which encrypts stored data at rest.
- All traffic to the app, the API and the MCP server uses HTTPS.
- Payments are handled by Stripe’s hosted checkout and customer portal. Card details never reach our servers.
- The full list of providers is on the Subprocessors page.
Secrets and credentials
- Customers’ SMTP passwords, Google OAuth refresh tokens and any third-party API keys they store are encrypted in the application with AES-256-GCM before being written to the database. The key is kept in the hosting environment, not in the database.
- AutoTalent API keys (for the MCP server, CLI and API) are shown once when created and stored only as a keyed hash (HMAC-SHA256). They can be created read-only and revoked at any time in Settings.
- Gmail connections ask only for permission to send (
gmail.send), not to read mail.
Authentication and tenant isolation
- Sign-in, sessions and organization membership are handled by Clerk. Every page and API route outside the public site requires a signed-in user who belongs to an organization.
- Only organization admins can change plans. Billing and data settings apply to the whole organization.
- Customer data is separated by organization. The application server scopes every query to the signed-in user’s organization, and automated tests check that scoping. Row Level Security is also enabled on the core tables, which blocks access through the database’s public API key. Because the server uses a privileged database connection, the application’s own scoping is the main control.
- Hiring-manager review links work without signing in. Each link is signed (HMAC), covers only the candidates the recruiter chose to share, and expires after 14 days.
Webhooks and email
- Webhooks from Stripe and Clerk are accepted only with a valid signature.
- Webhooks from SendGrid (inbound replies and delivery events) are checked against a shared secret.
- Outreach sent through SendGrid is signed with DKIM for the sending domain. Open and click tracking are turned off, so emails contain no tracking pixels or rewritten links.
Data handling
- AI providers receive the candidate and requisition data needed for each task (see AI & automated decisions).
- Application logs can contain names, email addresses and short previews of candidate replies while we debug delivery. They stay in our hosting providers’ logging, subject to those providers’ log retention.
- Deletion is currently handled on request by email rather than self-service. See Retention and deletion.
What we don't have yet
- No SOC 2, ISO 27001 or other third-party security certification or audit.
- No customer-facing audit log of user actions.
- No automatic data retention schedule, and no self-service data export or deletion for candidates.
- No single sign-on (SAML) for customer organizations.
If one of these blocks you from using AutoTalent, tell us. It helps us decide what to build next.
Reporting a vulnerability
Email support@autotalent.ai with “Security” in the subject. Include the steps to reproduce. Please don’t access other customers’ data, degrade the service or run automated scans that send email. We’ll acknowledge your report and keep you updated while we fix it.