Data teams ship web apps now
The line between data engineering and application development has dissolved. Data teams build and deploy dashboards, internal tools, customer-facing analytics portals, and API layers on top of warehouses. These are production web applications, exposed to the same threats as any SaaS product, yet most data teams treat security as someone else's problem.
The consequences are real. A misconfigured Supabase table leaks customer records. An unprotected admin panel exposes your entire data pipeline. A debug endpoint left in production hands an attacker your database connection string. These are not theoretical risks: they are the most common findings when data-driven applications are scanned with tools like Keybrum.
The 12-point security checklist
1. Authentication is not optional
Every route that reads or writes data needs authentication. This includes API endpoints, webhook receivers, admin panels, and queue dashboards. "It's internal" is not a security policy: internal tools get indexed by search engines, discovered by port scanners, and accessed by former employees.
Test it: try accessing every route without credentials. Keybrum's free security check does exactly this, testing over 300 attack patterns against your live application in minutes.
2. Row-level security on every table
If you use Supabase, Firebase, or any backend where the client talks directly to the database, row-level security (RLS) policies are your authorization layer. A missing policy on one table means any authenticated user can read every row in that table.
Audit every table. If RLS is disabled, you need a reason in writing. If a policy uses true as the check expression, you have no policy at all.
3. No secrets in client-side code
JavaScript bundles, source maps, and environment variables prefixed with NEXT_PUBLIC_ or VITE_ are visible to anyone who opens the browser console. API keys for OpenAI, Stripe, Twilio, or your data warehouse have no business there.
Scan your deployed application, not just your repository. A secret that was in the bundle three deploys ago is still in the CDN cache. Automated scanners like Keybrum check every JavaScript file, source map, and HTML page for leaked credentials, then verify whether the key is still active.
4. Lock down your admin routes
Bull Board, Redis Commander, Prisma Studio, Drizzle Studio, phpMyAdmin: every database or queue management UI that ships in your Docker image is a liability in production. If it responds to an unauthenticated request, it's a critical vulnerability.
Either remove these tools from your production build or protect them with authentication middleware. A /admin path prefix is not protection: it is an invitation.
5. CORS is not a firewall
Permissive CORS headers (Access-Control-Allow-Origin: * with credentials) let any website make authenticated requests to your API on behalf of your users. For a data application that serves analytics, this means any page on the internet can read your users' data.
Restrict allowed origins to your actual frontend domains. Reflect the Origin header only after checking it against an allowlist, never with a regex that matches too broadly.
6. Rate-limit everything that costs money
AI endpoints, warehouse queries, and third-party API calls cost real money per request. Without rate limiting, a single attacker can exhaust your OpenAI budget or run up your Snowflake compute bill in hours. Data teams are especially exposed because a single query can scan terabytes.
Build your architecture diagram now
Datadef generates professional diagrams with AI, 2,000+ cloud icons, column-level data lineage, and an MCP server your coding agent can drive.
The second six
7. Validate and sanitize all inputs
SQL injection is not a 2005 problem. It is the number-one database attack in 2026. If your application constructs queries from user input (even "just for filtering"), you need parameterized queries. No exceptions, no "we'll fix it later."
8. Keep dependencies current
The npm and pip ecosystems disclose 50+ vulnerabilities per week. A dependency you installed six months ago may now have a known exploit. Run npm audit and pip audit in CI, and block deploys that introduce critical vulnerabilities.
9. Encrypt data in transit and at rest
HTTPS is not enough. Database connections, internal service calls, and message queues should use TLS. Data at rest in object storage and warehouses should use server-side or client-managed encryption keys. Check that your TLS certificates are valid, your HSTS headers are set, and your cookies carry the Secure flag.
10. Log access, not just errors
When something goes wrong, you need to know who accessed what and when. Structured access logs with user identity, IP, timestamp, and resource accessed are the minimum. Ship them to a separate store that an attacker cannot delete after gaining access to the application.
11. Test from the outside
Internal code reviews and unit tests cannot catch what an attacker sees. You need to test your application the way a stranger would: from the public internet, with no credentials, following every link and probing every endpoint.
Keybrum's automated pentest does this in under ten minutes: it visits your application, reads the JavaScript, maps the backend, and runs 338 attack checks covering the OWASP Top 10 and beyond. The report shows exactly what leaked, with proof and fix instructions you can paste into your IDE.
12. Document your architecture
You cannot secure what you cannot see. A data platform diagram that shows every service, database, queue, and external API is the first step in threat modeling. It reveals which components are internet-facing, which hold PII, and which connections cross trust boundaries.
Tools like Datadef make this easy: generate your architecture diagram from your repository, track data lineage, and keep the documentation in sync as your platform evolves.
Automate the audit
Manual security reviews do not scale. A data team shipping weekly cannot afford a quarterly pentest as its only security gate. The modern approach: run automated security scans on every deploy, the same way you run lint and tests in CI.
Keybrum starts at €49 per application and runs in CI via its CLI and MCP server. Every deploy gets a security report, every new vulnerability gets flagged before it reaches production. For data teams building on Supabase, Firebase, or Next.js, it covers the specific attack surface these platforms create: RLS bypasses, exposed storage buckets, leaked API keys, and unprotected server actions.
Security is not a feature you add at the end. It is a constraint you build with from the start. This checklist is the minimum. Run it, automate it, and ship with confidence.