A working feature can cross the wrong boundary
Consider an illustrative change to a customer portal. An AI coding assistant adds an invoice-download endpoint. It checks that the caller is signed in, looks up an invoice by its ID, and returns the document. A test confirms that a customer can download their own invoice.
The unanswered question is whether that customer can download another organization’s invoice. Authentication establishes who the caller is. Authorization must establish whether that caller may perform this action on this record.
OWASP’s guidance on broken object-level authorization describes this class of failure and calls for permission checks on endpoints that act on objects identified by the request. The risk exists regardless of who wrote the code. AI-assisted implementation makes it especially useful to state the boundary explicitly instead of assuming it was understood.
For a software buyer, the consequence is concrete: a seemingly small feature can put customer data and customer trust at risk. A clean-looking diff is not enough evidence to settle that concern.
Give the reviewer your actual security expectations
A generic instruction to “check security” leaves too much interpretation. A useful project rule describes the behavior the system must preserve. For the invoice example: the caller must be authorized to download the requested invoice within the correct tenant, using the project’s approved authorization path.
Peertic’s repository and per-file rules let teams express expectations for the code being reviewed. Operators can inspect a file path to see the exact rule and source layer the reviewer will use. That makes the review policy something a security lead and an engineer can examine together.
The rule still needs sound design. Tenant membership alone may not grant access to every invoice or action. Your team should define the permissions that apply and provide the relevant context.
Look beyond the changed handler
The authorization check may live in middleware, a shared service, or a data-access layer. Reviewing only the new handler can miss a protection that exists elsewhere—or assume a protection that the handler never invokes.
Peertic’s review engine supports related-file bundles and codebase context. Linked requirements can explain intended behavior, while configured project rules guide the review. When a run raises a concern, line-level GitHub findings give engineers a specific location and explanation to investigate.
These capabilities provide a workflow for examining security-sensitive changes. They do not establish guaranteed detection of authorization flaws, and this invoice scenario is an evaluation example rather than a recorded Peertic result.
Turn a concern into evidence before merge
For this example, a useful review exercise would include:
- An allowed request: a caller with the required permission can download the intended invoice.
- A denied request: a caller from another tenant cannot retrieve that invoice by changing its identifier.
- A restricted caller: a user in the same tenant without the required permission is denied.
- A follow-up after the fix: inspect the updated authorization path, run the tests, and revisit the review finding.
Peertic can publish findings and configured merge checks, then run follow-up reviews. Keep specialist security checks and the responsible engineer involved in the decision. A successful AI review cannot prove the absence of vulnerabilities.
Evaluate on the boundaries your customers depend on
Bring a small set of changes involving your actual permissions model, including correct implementations and known defects. Have a qualified reviewer establish the expected outcomes, then assess what Peertic finds, what it misses, and how much investigation each finding requires.
The buying question becomes specific: can this review workflow help your team examine important access boundaries before merge, with useful evidence in the pull request? That is a more meaningful trial than asking for a generic security score.
Explore Peertic
Bring one permission boundary your team must protect.
Tell us how your application handles authorization and where its review rules live. We can discuss an evaluation using representative changes and your existing security checks.
Discuss security-focused review