Security
How we protect your data.
You trust FlashTower with a map of your AI risk. Here is exactly how we protect it — and what we have not done yet.
Controls
Security controls in place today
This list describes controls that exist in the product now. We update it as we add more.
Tenant isolation enforced in the database
Every table holding customer data carries an organization identifier and is protected by Postgres row-level security. Isolation is enforced on every query by the database itself, not only by the application or interface. We test cross-tenant access automatically.
Role-based access control
Organizations assign Owner, Admin, Security Admin, Analyst, Viewer, and Billing roles. Permissions are checked on the server for every action and mirrored in database policies. The last owner of an organization can never be removed or demoted.
Append-only audit log
Security-relevant actions — sign-ins, team and role changes, assessment completion, finding status changes, report generation, billing changes — are recorded in an append-only audit log: database triggers block edits and deletes, so entries cannot be edited or deleted through the application. Entries are kept for at least one year, then removed under the retention schedule (or when the organization is deleted). Secrets are never written to it.
Strong authentication
Accounts use email verification and hashed passwords managed by our authentication provider. Multi-factor authentication (TOTP) is available to every user and required for FlashTower staff with administrative access.
Encryption in transit and at rest
All traffic is served over HTTPS with HSTS. Data at rest is encrypted by our infrastructure providers (Supabase for the database and file storage, Vercel for application hosting). Evidence files live in a private bucket and are accessed only through short-lived signed URLs.
Data minimization
FlashTower assesses AI systems from metadata you provide — what a system can access and do — not from the underlying data. The anonymous free scan stores only aggregate results; your answers stay in your browser until you save them to an account.
No AI training on customer data
Risk scores are produced by deterministic rules, not AI models. We do not use customer data to train AI models. Optional AI-written explanations are off by default and only run when an organization explicitly enables them.
Hardened application
A Content Security Policy on every page (with per-request nonces and no inline scripts in the customer app and admin console), frame-ancestor denial, input validation on every endpoint, rate limiting, same-origin checks on state-changing requests, safe redirect handling, SSRF protections for outbound requests, and upload type verification.
Privacy-preserving monitoring
Error monitoring strips request bodies, cookies, authorization headers, and personal data before events leave our servers. Product analytics use allow-listed, non-identifying properties only — never assessment answers or report contents.
Certifications
Where we are — honestly
FlashTower does not yet have a SOC 2 report, is not certified under ISO/IEC 27001, and has not yet completed a third-party penetration test. We publish status updates on our trust center as milestones are reached. We will never display a certification or badge we have not earned.
Subprocessors
We use a small number of infrastructure providers for hosting, database, email delivery, payments, error monitoring, and product analytics. The current list is published on our subprocessors page.
Data retention and deletion
Each organization can configure its data-retention period and request export or deletion of its data. See our privacy policy for details.
What FlashTower does — and does not do
Observe and report. You stay in control.
FlashTower discovers AI systems, assesses what they can access, detects changes, and alerts your team with recommended steps. It does not change your systems on its own. Response actions run only through FlashTower Guard, only after your organization's own owners, admins, or security admins approve them or under an automation policy your owners or admins explicitly authorize — and FlashTower staff can never approve them for you. Optional integrations are listed in the app with their current availability; through them FlashTower only reads, and it requests the narrowest scope each provider offers. Guard is off and in dry-run by default and needs a separate write grant from you.
Limits of detection
FlashTower is designed to help identify, monitor, detect, assess, and report certain AI-related risks and security events. No security system can identify, detect, prevent, or report every threat, vulnerability, unauthorized action, malicious actor, AI behavior, configuration issue, or security incident.
- FlashTower findings and alerts are informational.
- FlashTower does not guarantee detection of every threat.
- FlashTower does not guarantee prevention of incidents.
- FlashTower does not certify compliance.
- FlashTower does not provide legal advice.
Found a vulnerability?
We welcome reports from security researchers and will not pursue legal action against good-faith research that follows our responsible disclosure policy. Please do not access other customers' data, degrade service, or use social engineering.
Describing our security controls is not a guarantee that FlashTower is free of vulnerabilities. No system is perfectly secure; we work continuously to find and fix weaknesses and will notify affected customers of security incidents as described in our Security Incident Terms.
See what your AI can access.
Run the free FlashTower Scan. Get your AI risk score and every finding in about 10 minutes — no account or system access required.