A researcher has released a static analysis module designed to detect IDOR (Insecure Direct Object Reference) vulnerabilities in Python web applications by analyzing data flow rather than relying on heuristics.
The project addresses a long-standing question: whether IDOR flaws can be identified statically by tracing the connection between a user-supplied identifier, the retrieved object, and any authorization check that must dominate the access path.
IDOR occurs when an application exposes direct references to objects without verifying that the current user is authorized to access them. Classic examples include Flask routes that accept an invoice_id parameter and fetch the record with Invoice.query.get(invoice_id) after only a @login_required decorator, allowing any authenticated user to enumerate other users’ invoices.
The new module parses project sources to locate HTTP handlers across Django, DRF, Flask, and FastAPI. It then builds a protocol of inputs, data accesses, and checks for each handler. Values are labeled using the SIAOD system: SUBJECT for the authenticated user, INPUT for ordinary request data, ATTACKER_SELECTED for identifiers that select specific objects, OBJECT for records retrieved from storage, and DYNAMIC for values whose origin cannot be determined statically.
Unlike conventional taint analysis, the engine requires that an authorization check both references the exact OBJECT reached via an ATTACKER_SELECTED identifier and dominates the code path that returns or uses that object. Checks in unrelated branches, checks performed after the object has already been returned, or checks that compare a different object are all rejected.
Evaluation on 150 small open-source repositories produced 112 findings, of which 48 were confirmed true positives. On 15 production codebases totaling 12 million lines, the tool reported 731 findings but only five were genuine; the remainder stemmed from authorization logic implemented in routers, dependencies, or external RBAC frameworks rather than inside individual handlers.
The author concludes that while mathematical tracking of object ownership is possible, real-world authorization patterns frequently place the decisive checks outside the analyzed handler scope, limiting the precision of purely static approaches today.
