Starter
Try AI PR review.
- 2 reviewers
- Code Review
- Security Review
- 3 seats included
- 15,000 credits to start, then 5,000/mo
- About 70 reviews to start, then about 25 a month
- Unlimited repos
Finds auth gaps, injection and exposed secrets in changed code.
MediumKnown issue in the Security Review benchmark · caught by DeepSeek V4.1 FlashSee the graded case
Real IonWarp comments on open-source pull requests, graded against the benchmark's answer key.
Missed byQodo 3Greptile 5CodeRabbit 3Cursor Bugbot 6
3 of 9 · See the benchmark
P1Blocked-email list is matched case-sensitively — trivially bypassed at signup · Security Review
When an admin blocks an email, a blocked spammer re-registers with different letter case ('BAD@spamclub.com') and the account is created: should_block? does a case-sensitive where(email:) and the signup path never downcases the submitted email, so the new control fails its one job.
CriticalKnown issue in the Security Review benchmark · caught by GLM 5.3 FlashSee the graded case
P2Single-use guarantee of backup codes is not concurrency-safe (TOCTOU) · Security Review
When two login requests supply the same backup code concurrently, both read the stored array before either write lands, so both pass indexOf and both log in. This breaks the UI promise in the new locale string backup_code_instructions: 'Each backup code can be used exactly once'. Trigger: parallel signIn calls with the same code (e.g. double-clicked submit).
HighKnown issue in the Security Review benchmark · caught by GLM 5.3 FlashSee the graded case
P1Local cache is set before CreateOrUpdateDevice, so a limit-rejected device is silently whitelisted for 29 minutes · Security Review
When the device limit is reached, tagDeviceUI caches the device key before CreateOrUpdateDevice returns ErrDeviceLimitReached; the failed request errors for the anonymous user (client.go:46), but the next request from the same device hits the cache, returns nil, and proceeds untagged — one failed load then a limit bypass per node per 29min.
HighKnown issue in the Security Review benchmark · caught by GLM 5.3 FlashSee the graded case
GLM 5.3 Flash is its best model on this benchmark; Security Review runs DeepSeek V4.1 Flash unless you pick another.
IonWarp's models and the review apps, graded on the same open-source pull requests.
| PRs | |||||
|---|---|---|---|---|---|
| IonWarp Security · GLM 5.3 Flash | |||||
| IonWarp Security · DeepSeek V4.1 Flash | |||||
| Qodo v2 | |||||
| IonWarp Security · Muse Spark 1.3 | |||||
| Greptile v4.1 | |||||
| IonWarp Security · Grok 4.6 | |||||
| IonWarp Security · Gemini 3.8 Flash | |||||
| CodeRabbit | |||||
| Cursor Bugbot | |||||
| IonWarp Security · DeepSeek V4 Pro | |||||
| IonWarp Security · GPT 5.6 Luna Pro |
On this benchmark GLM 5.3 Flash scored higher than DeepSeek V4.1 Flash (66.7% F1 against 62.5%). Why IonWarp runs DeepSeek V4.1 Flash: in a trial of our specialist reviewers on production pull requests on 23 September 2026 it filed about as many findings per review as GLM 5.3 Flash at high effort (0.93 against 1.02) for a third of the cost ($0.039 against $0.125), and 3 of its 64 findings were refuted. The benchmark ran GLM 5.3 Flash at low effort.
One workspace price with seats included. Credits pool across your team. Unlimited repos on every plan.
Security Review is on every plan, including Starter.
Try AI PR review.
Get AI PR review right from day 1.
Scale AI PR review.
Custom reviewers + credits · GitHub Enterprise · SSO/SAML + audit logs · Custom security review + SLA
P2Email-domain whitelist is matched without an anchor, so a suffix-domain address bypasses the registration allowlist · Security Review
When `email_domains_whitelist` is set, `email_in_restriction_setting?` builds `Regexp.new("@(#{domains})", true)` and tests it against the caller-supplied signup email with `=~`. The pattern is not anchored, so it is a substring match: an anonymous caller on a site that only admits `@company.com` can register `attacker@company.com.evil.tld`, because that string contains `@company.com` and therefore satisfies the whitelist. The same unanchored pattern wrongly rejects `notgood@mailinator.com.evil.tld` on the blacklist branch. The value reaches the check straight from the request: `app/models/user.rb:87 user.email = params[:email]` via the public, CSRF-exempt `POST /users` create action. Actor: unauthenticated signup caller; gain: an account on a whitelist-restricted forum. Note the sibling dot-escaping shows the intent is a real domain match, and the whitelist is the only gate here.