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
OpenAI's Codex Security scans a repository for vulnerabilities. IonWarp reviews pull requests for the security holes they add.
MediumKnown issue in the Security Review benchmark · caught by DeepSeek V4.1 FlashSee the graded case
Codex Security is OpenAI's application security agent.[1] It scans a repository, ranks likely vulnerabilities and tries to reproduce each one in a clean container; you review any patch before it becomes a draft pull request.[2]
IonWarp's Security Review reads one pull request before it merges. On 13 benchmark pull requests, DeepSeek V4.1 Flash, the model it runs, found 10 of 18 known security issues for about 6 cents a review; the Qodo app scored highest, an F1 of 73.3% to 62.5%.[3] What a security code review checks. Anthropic's pull request check is Claude Code security review.
In Sentry #67876 the GitHub OAuth state was the pipeline's fixed signature, the same for every user, so the CSRF check did nothing. DeepSeek V4.1 Flash flagged it. Bugbot, CodeRabbit and Greptile missed it, and so did GLM 5.3 Flash, the benchmark's top model.[4]
Security ReviewOn Starter, Pro and Max
OpenAI facts are from its Codex Security documentation, read on 2026-10-04; numbered notes link to each page.
| Criterion | IonWarp | Codex Security |
|---|---|---|
| What it is | A security reviewer in a GitHub app; the planner adds it to the pull requests that need it | An application security agent that helps teams find, confirm and fix vulnerabilities[1] |
| What it reads | A pull request's changes and the code around them | A repository or a folder, a pull request or branch before merge, or the new commits it monitors[1][2] |
| What it returns | Findings on the pull request with a severity, the attack path, a suggested fix and an AI fix prompt | Ranked findings with criticality, validation evidence, remediation guidance and a proposed patch when one is available[2] |
| Where it runs | IonWarp's hosted service | The Codex app, the open-source CLI and SDK, or Codex cloud for connected GitHub repositories[1][5] |
| Model | F1 | Precision | Recall | $ per review | Median time |
|---|---|---|---|---|---|
| DeepSeek V4.1 FlashThe model it runs | 62.5% | 71.4% | 55.6% | $0.06 | 5m 56s |
| GLM 5.3 FlashA failed review retries on it · best on this benchmark | 66.7% | 83.3% | 55.6% | $0.01 | 1m 38s |
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.
Measured on the same 13 open-source pull requests, with known issues. See the benchmark
Real IonWarp comments on open-source pull requests, graded against the benchmark's answer key.
Missed byQodo 3Greptile 5CodeRabbit 3Cursor Bugbot 6
2 of 9 · See the benchmark
P2Migration interpolates site_settings value directly into INSERT SQL · Security Review
When the legacy 'embeddable_hosts' site_setting value contains a single quote or SQL metacharacter, `execute "INSERT ... VALUES ('#{h}', ..."` builds that value into raw SQL. The value is admin-controlled (not an untrusted actor), so no privilege gain, but a stray quote corrupts or fails the migration on upgrade.
CriticalKnown issue in the Security Review benchmark · caught by GLM 5.3 FlashSee the graded case
P2Blacklisted-guest filter is case-sensitive on the guest email · Security Review
When BLACKLISTED_GUEST_EMAILS is set, `!blacklistedGuestEmails.includes(guest)` compares the raw caller-supplied email against lowercased blacklist entries, so 'Evil@X.com' bypasses 'evil@x.com' and still receives the booking ICS with event details.
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.
One workspace price with seats included. Credits pool across your team. Unlimited repos on every plan.
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
OpenAI's application security agent. It scans a repository, builds a threat model, tries to reproduce each likely issue in a clean container, and proposes patches that you review before opening a draft pull request.
No. Codex code review posts a review on a pull request when you ask for one or turn on automatic reviews. Codex Security is a separate security agent that scans repositories and commits and tries to reproduce what it finds.
Yes. Codex Security scans and validates across the repository; IonWarp's Security Review comments on the pull requests whose changes need it. Both work on the same GitHub repositories.
Security Review runs on Starter, Pro and Max. Starter is free for 3 seats with 15,000 credits on signup.
P2OAuth `state` is the deterministic pipeline signature, so the callback's CSRF binding is void · Security Review
When the user is sent to GitHub, `state` is set to `pipeline.signature`, which is `md5_text(*pipe_ids).hexdigest()` over the pipeline view class names (sentry/pipeline/base.py). It is a public constant — the PR's own test hard-codes it (`state=9cae5e88803f35ed7970fc131e6e65d3`). On the callback (`.../extensions/github/setup/`, served by PipelineAdvancerView with `auth_required=False, csrf_protect=False`) the check `request.GET.get("state") != pipeline.signature` therefore always passes for anyone who computes the constant. The established sibling, `OAuth2CallbackView`, instead uses `secrets.token_hex()` bound into pipeline state and compares against `pipeline.fetch_state("state")`. Consequence: an attacker completes GitHub's authorize step for their own account to obtain a `code`, then gets a victim who has an in-progress GitHub integration pipeline to load `.../extensions/github/setup/?code=<attacker code>&state=9cae5e88...&installation_id=<attacker installation>`. The victim's pipeline binds `github_authenticated_user` to the attacker, and because the supplied installation's `metadata["sender"]["login"]` is also the attacker, the new match check passes and the attacker's GitHub installation is linked into the victim's Sentry organization (login/CSRF on the OAuth callback).