The Novee Gym is open for business at Black Hat Europe

Book a meeting

The Novee Gym is open for business at Black Hat Europe

Book a meeting

IDOR Vulnerability: Real-World Examples and How to Catch It

Key Takeaways A pentest flagged the flaw that exposed 800 million First American title documents months before a journalist reported it to the company. That flaw was an insecure direct object reference (IDOR) vulnerability, and exploiting it took nothing more than editing a link. CISA, the NSA, and the Australian Cyber Security Centre warn that […]

Novee Team

14 mins

Explore Article +

Key Takeaways

  • IDOR is an ownership check the server skipped: Changing one identifier in a URL, request body, or API call returns another user’s record whenever the application fails to tie the requested object to the person requesting it.
  • Real breaches follow the same simple pattern: First American, Panera, USPS, McHire, India’s tax portal, and Vetco exposed millions of records through identifiers anyone could change, and several flaws stayed open long after someone reported them.
  • Detection needs more than one identity: IDOR surfaces only when testing compares what different users and tenants can reach, so authorization testing has to run with multiple accounts on every release.

A pentest flagged the flaw that exposed 800 million First American title documents months before a journalist reported it to the company.

That flaw was an insecure direct object reference (IDOR) vulnerability, and exploiting it took nothing more than editing a link. CISA, the NSA, and the Australian Cyber Security Centre warn that IDOR flaws are frequently exploited because they are common, hard to prevent outside the development process, and easy to abuse at scale. 

Every new endpoint that accepts an object ID can repeat the same mistake, and a pentest once or twice a year sees only a slice of those endpoints.

Teams that catch IDOR reliably test authorization the way attackers probe it, with multiple users against every endpoint that references an object, every time the code changes.

The incidents that follow show which check failed each time, how the IDOR and BOLA definitions shape test scope, and the method that finds these flaws first.

What IDOR Is and Why It Tops Every API Authorization Risk List

Insecure direct object reference describes an access control failure where an application receives an identifier from the client, such as an invoice number, user ID, or file name, and uses it to fetch a record. The reference itself is normal, but the vulnerability appears when the server returns the record without checking that the requester owns it.

An IDOR attack can be as small as a one-digit edit to a request the attacker is already allowed to send.

GET /api/invoices/10432    200 OK   (attacker’s own invoice)

GET /api/invoices/10433    200 OK   (another customer’s invoice)

The joint advisory from CISA, the NSA, and ACSC groups these flaws into four types. Horizontal IDOR is a form of horizontal privilege escalation, where a user reaches another user’s data at the same privilege level. Vertical IDOR reaches data reserved for higher roles, and object-level and function-level IDOR let users modify objects or run actions they should never touch.

In the latest OWASP Top 10, every application tested had some form of broken access control, a category that names IDOR directly. The OWASP API Security Top 10 ranks broken object-level authorization as the first API risk, because APIs take object IDs from the client on nearly every call. Those same APIs often serve mobile clients, which is why mobile app testing has to reach the API layer behind the app.

6 Real IDOR Breaches (And the Gaps That Made Them Possible)

Each incident below traces to one missing ownership check. The incidents differ in who could exploit the flaw and how long it stayed open, and several were reported well before anyone fixed them.

First American Financial: Document Links Anyone Could Edit

First American’s EaglePro application shared title and escrow documents with customers through web links. Changing digits in one of those links returned documents belonging to other customers. 

The SEC order states that the repository held roughly 800 million document images dating back to 2003, and that a January 2019 manual penetration test had already identified the vulnerability. Many of the images contained Social Security numbers and bank records, and the flaw stayed unremediated until a journalist exposed it in May 2019.

The consequences arrived in stages: 

  1. The SEC charged the company with disclosure control failures and imposed a $487,616 penalty in 2021. 
  2. New York’s Department of Financial Services added a $1 million penalty in 2023 for cybersecurity regulation violations tied to the breach.

Document retrieval had no authorization check, so a single link pattern opened the entire repository.

Panera Bread: Eight Months of Warnings

In August 2017, a researcher told Panera that its delivery site exposed customer records through an API. Customer accounts used sequential integer IDs, so anyone could step through them and collect names, email addresses, home addresses, birthdays, and the last four digits of saved cards. 

Panera said it was working on a fix. Eight months later, the site was still leaking customer records in plain text, and Panera took it offline only after KrebsOnSecurity contacted the company.

Panera initially put the exposure at about 10,000 customers. Hold Security’s analysis put the figure at roughly 37 million records, and the incident became a case study in how a slow response magnifies a simple flaw.

The API required no authentication and performed no ownership check, and sequential IDs turned that gap into a full database export.

USPS Informed Visibility: Logged In Meant Authorized

The U.S. Postal Service’s Informed Visibility API gave business mailers near real-time tracking data. It also let any logged-in usps.com user query account details belonging to other users, including email addresses, usernames, account numbers, street addresses, and phone numbers. Many API functions accepted wildcard search parameters, so one query could return every record in a data set, and some account details could be modified.

The exposure touched about 60 million users. A researcher reported the flaw to USPS more than a year before the fix, and the agency acted only after KrebsOnSecurity contacted it.

The only access control on those API calls was an active session, so every authenticated account could read every other account.

McDonald’s McHire: One Decremented Lead ID

McHire is the applicant platform McDonald’s franchisees use to screen job candidates. In June 2025, two researchers signed into a staff login for Paradox, the vendor behind McHire, using the default credentials 123456, which opened a test restaurant account. 

While reviewing their own test application, they found an internal API that fetched candidate records by lead_id. Decrementing that number returned another applicant’s personal data, and the ID range pointed to more than 64 million applications.

Each response included names, email addresses, phone numbers, home addresses, and an auth token that logged into that applicant’s session and exposed their chat history. Paradox confirmed the issues were resolved about a day after disclosure.

The API accepted any authenticated session and never checked whether a lead belonged to the caller’s restaurant, and the leaked tokens turned data exposure into account takeover.

India’s Income Tax Portal: Swapping One Taxpayer ID for Another

India’s e-Filing portal lets taxpayers sign in with their Permanent Account Number (PAN) to view their records. In September 2025, two researchers found that swapping their own PAN for someone else’s in the page’s network request returned that person’s sensitive data. The response included full names, home addresses, dates of birth, phone numbers, bank account details, and Aadhaar numbers, the national identifier used across government services.

The flaw also exposed people who had not yet filed returns that year. The portal lists more than 135 million registered users, and it remains unclear how long the bug existed or whether anyone else used it. The tax department fixed it after the researchers alerted India’s CERT-In.

The server confirmed a user was logged in and then trusted whatever PAN the browser sent, a textbook case of horizontal escalation.

Petco’s Vetco Clinics: Sequential Customer Numbers on a Public Page

Vetco’s customer portal generated PDF copies of veterinary records. The PDF endpoint required no login and took a customer number in the web address. Because customer numbers were sequential, changing one or two digits pulled another customer’s files, and Google had already indexed at least one record.

The files held visit summaries, medical histories, prescription and vaccination records, owner names, home addresses, email addresses, phone numbers, and signatures. Sampling suggested millions of customers could have been exposed, with records dating to at least mid-2020. Petco took part of the site offline in December 2025 and would not say whether logs could show if data was taken. TechCrunch counted it as Petco’s third data security incident of the year.

A document generator was reachable without authentication and keyed on guessable numbers, leaving nothing standing between the internet and customer records.

How IDOR and BOLA Differ and What That Means for Testing

Both terms describe overlapping failures, and security practitioners often use them interchangeably. 

But there are clear distinctions to be made. MITRE maps both to CWE-639 and notes that OWASP’s IDOR term is broader, while BOLA may apply to other weaknesses as well. A BOLA vulnerability is the API-era framing, centered on whether every endpoint that accepts an object ID enforces authorization for that specific object and action.

Each term points testers toward a different part of the attack surface. Here’s how the differences show up in scope:

DimensionIDORBOLA
OriginOWASP Top 10 for web applicationsOWASP API Security Top 10
ScopeObject references in URLs, forms, cookies, and files, including path traversal in OWASP’s framingObject-level authorization in API endpoints, REST and GraphQL
Where IDs appearQuery strings, form fields, file pathsPath parameters, JSON bodies, headers, GraphQL arguments
Typical test surfaceBrowser workflows and page requestsEvery API operation that reads, updates, or deletes an object
Detection implicationMap user-facing parameters that reference objectsEnumerate every endpoint and method, including undocumented ones

Scoping tests to the BOLA definition pushes coverage into every PUT, PATCH, and DELETE operation, where a missing check lets one user change or erase another user’s data. It also forces an inventory of undocumented endpoints, where exposures like Vetco’s public PDF generator tend to hide.

Why Most Automated Scanners Cannot Detect IDOR

A successful IDOR attack returns an ordinary response. The status code is 200, the JSON is valid, and nothing in the payload signals that the record belongs to someone else. 

The joint government advisory on IDOR states that not all DAST tools can detect these flaws, and tools that can may need specific environment configuration and custom rules while still falling short of foolproof.

Four structural limits explain why scanning from a single session misses most of these flaws:

  • One session, one perspective: A scanner authenticated as a single user has no second identity to compare against, so it cannot determine whether a returned object belongs to the requester.
  • No map of ownership: Detecting an IDOR vulnerability requires knowing which roles and tenants should access which objects. That knowledge resides in the application’s business logic, and pattern-based scanning never builds it.
  • Identifiers hide deep in workflows: Object IDs appear in JSON bodies, headers, GraphQL arguments, and multi-step flows that crawlers rarely reach in a valid state.
  • Write-side checks need controlled data: Proving that one user can modify or delete another user’s object requires seeded test records, and generic scanners avoid destructive requests by design.

Skilled pentesters catch IDOR with multiple accounts and application knowledge, but their coverage ends when the engagement time runs out.

How to Test for IDOR Across Web and API Environments

OWASP’s testing guide sets a clear baseline for authorization testing, and structured penetration testing frameworks turn that baseline into a repeatable plan. The method applies equally to server-rendered pages, REST APIs, GraphQL, and the APIs behind mobile clients.

Provision Multiple Test Identities

OWASP recommends testing with at least two users who own different objects, plus users with different privilege levels. Add accounts in separate tenants for multi-tenant applications. Seed each account with records you can recognize, so a leaked object stands out immediately in a response.

Map Every Object Reference

Proxy traffic while each user walks through the application, and import API specifications where they exist. Record every parameter that names an object, including path segments, query strings, JSON fields, headers, cookies, and GraphQL arguments. Note identifiers that appear in responses, since those hand an attacker the next ID to request.

Swap References Across Users and Methods

Replay User A’s requests with User B’s session and object IDs, then reverse the roles. OWASP’s API testing guidance applies the same swap to REST and GraphQL endpoints. 

Test GET, PUT, PATCH, and DELETE separately, because read and write paths often run different code. Run lower-privilege sessions against admin-owned objects to catch vertical cases, and try wrapped, encoded, or array-formatted IDs to expose inconsistent parsing.

Validate Impact and Retest the Fix

Confirm that returned data belongs to the other account, capture reproducible steps, and rate severity by data sensitivity and write access. After the fix is deployed, replay the original request with the same identities. 

Add cross-user checks to CI so new endpoints are tested as they ship, since each release can introduce a new BOLA vulnerability.

Catch IDOR Before It Ends Up in a Regulator’s Filing

Every incident above stayed open for months or years, and several stayed open after someone report them. 

First American’s own pentest found the flaw long before a journalist forced the fix, showing that point-in-time testing and unverified fixes leave that window open on every release.

Novee closes that window with continuous AI pentesting across web applications, APIs, and mobile apps. Its multi-model AI offensive system builds an Asset Intelligence Model of each application’s workflows, roles, permissions, and APIs, then tests whether one user or tenant can reach another’s objects. 

The Research Agent probes authorization gaps like IDOR, BOLA, and BFLA. The Validation Agent confirms every finding with a working exploit and reproduction steps, and the Remediation Agent delivers stack-specific fixes and retests the original request to confirm the check holds.

Book a demo today to see how Novee finds and validates IDOR and BOLA flaws across your web applications and APIs.

FAQs

Is IDOR always exploitable regardless of the object type?

Exploitability depends on whether the server skips the ownership check, and impact depends on the object. Public objects, such as published product pages, carry no IDOR risk. Read access to a display name is low severity, and write or delete access to invoices, credentials, or session tokens is critical. References can be numeric IDs, strings, file paths, or composite keys split across parameters.

Can IDOR exist in APIs that use UUIDs?

Yes. UUIDs make identifiers harder to guess, and they add no authorization check. They leak through API responses, shared links, logs, emails, and related endpoints, and time-based UUID versions are partly predictable. Once an attacker holds a valid UUID, a missing ownership check exposes the object exactly as a sequential ID would. Treat random identifiers as defense in depth.

How does horizontal privilege escalation differ from vertical in IDOR?

Horizontal escalation happens at the same privilege level, such as one customer reading another customer’s orders. Vertical escalation happens when a lower-privilege user reaches objects or functions reserved for higher roles, like a standard user opening an admin export. Vertical cases often overlap with broken function-level authorization. Horizontal flaws appear more often, and vertical flaws tend to carry higher impact.

Can IDOR be exploited without valid authentication?

Yes, whenever an endpoint serves objects by ID and requires no session. Common examples include share links, document download and PDF generation endpoints, and legacy APIs. Unauthenticated IDOR is especially dangerous because anyone on the internet can enumerate records, search engines can index them, and exploitation leaves no account trail to investigate.

What is the remediation for an IDOR vulnerability?

Enforce an ownership or tenant check on the server for every request that reads, changes, or deletes an object, tied to the authenticated session. Deny access by default and centralize authorization logic in shared middleware or policy code. Ignore owner and tenant IDs sent by the client. Log spikes in denied requests, then retest with multiple accounts and add those tests to CI.

Stay updated

Get the latest insights on AI, cybersecurity, and continuous pentesting delivered to your inbox