Research · Application Security

Finding Authorization Flaws: IDOR and Access Control Testing in Web Applications

14 min read By Sushil Ojha
OWASPBurp SuiteIDORAccess Control

Authentication asks “who are you?” Authorization asks “what are you allowed to do?” Confusing the two is how applications end up with perfect login pages and porous object access. Broken access control — still perched near the top of OWASP rankings for good reason — remains one of the most common and expensive classes of web flaws because it often looks like a feature until someone else’s data appears in your session.

This write-up is a field method for finding horizontal and vertical privilege problems and insecure direct object references (IDOR) without relying on luck, folklore, or a single scanner checkbox.

Why authorization bugs survive “secure” teams

  • UI hiding is mistaken for enforcement
  • Server trusts client-supplied IDs, roles, or tenant hints
  • Multi-step workflows authorize the first hop and freestyle the rest
  • APIs multiply faster than permission matrices
  • Test accounts exist, but nobody systematically swaps them mid-request

Scanners help with plumbing. Authorization needs adversarial empathy: think like a user who is slightly more curious than polite.

Build a permission map before you click randomly

List roles (anonymous, user, admin, support, tenant-admin). List objects (orders, documents, messages, admin configs). List operations (read, create, update, delete, export, share). That matrix becomes your test plan. If the product team cannot describe intended access, you have already found a design defect — testing will only make it loud.

Horizontal escalation and IDOR

Authenticate as User A. Capture requests that reference object IDs. Replay as User B — or simply change the ID while remaining User A — and watch:

  • Numeric IDs, UUIDs, filenames, and secondary keys in paths, queries, bodies, and headers
  • Export, share, download, and “preview” endpoints that skip the same checks as the main UI
  • Batch endpoints that accept arrays of IDs and return mixed ownership results

Success criteria for a finding: reproducible access to another subject’s object without an intended relationship. Screenshots and request/response pairs beat adjectives.

Vertical escalation

  • Access /admin, role management, or privileged APIs directly as a low-privilege user
  • Tamper with role claims in JWTs, hidden fields, or request bodies — then confirm the server re-validates
  • Test “support” or “impersonation” features for unintended persistence of elevated context

If the only thing stopping admin access is a front-end route guard, congratulate the designer and file a critical.

A Burp-centered workflow that scales

  1. Create at least two users per role boundary you care about
  2. Browse features organically; let history and site map fill
  3. Use match/replace or auth macros carefully when swapping sessions
  4. Replay object-ID requests across users; note status codes and body leakage
  5. Automate only after you understand the patterns — blind fuzzing wastes trust

Signals beyond HTTP 200

Authorization bugs hide behind 200 responses with empty arrays, 403s that still leak existence, GraphQL errors that reveal schema, and verbose “access denied” messages that confirm object presence. Read bodies. Compare content lengths. Check exports. The status code is a rumor; the payload is testimony.

Reporting so engineering can fix it

  • Impact — whose data, what operation, what blast radius
  • Reproduction — exact steps, two accounts, request samples
  • Root cause hint — missing server-side check, IDOR on secondary key, broken tenant isolation
  • Fix — authorize on every request against the authenticated principal; never trust client role hints; centralize checks where possible

Prevention habits worth defending in design review

  • Server-side authorization for every object access
  • Indirect references or capability tokens where practical — still authorize
  • Automated tests that attempt cross-user access in CI
  • Deny-by-default for new endpoints; permissions as code reviews, not afterthoughts

Closing thought

Authorization testing is disciplined suspicion applied with two accounts and a proxy. Map intended permissions, cross the boundaries on purpose, and demand server-side proof. The teams that get this right treat access control as a core product property — not a UI courtesy.

← Work & Write-ups