Skip to main content
Cantina

Closes in

0 hours and 0 minutes left

We hack you first

Register below to receive a coupon.

Apex found bugs in

  • GitLab
  • Anthropic
  • Apple
  • Coinbase
  • Sentry
  • Supabase
  • Sierra AI
  • TikTok
  • Superhuman
01 Finding detail

Every finding tells the full story

Summary, affected code, a recorded proof of concept, and a fix checklist in one view. Validate it, comment on it, or hand it straight to a fix review.

  • Impact context on every issue
  • PoC code with highlighted lines and terminal replay
  • Fix-ready recommendations
Crit OPEN-3 Valid Merged

Spoofed Control UI locality silently mints persistent admin device tokens

OpenClaw audit openclaw/openclaw @ main Confidence 0.99

Summary

Same-LAN or shared-token caller can spoof Control UI locality, silently pair an admin device, and retain admin authority after shared-token rotation.

Proof of concept

const config = { gateway: { bind: "lan", auth: { mode: "token", token }, controlUi: { enabled: true, allowedOrigins: [`http://localhost:${port}`], }, }, };

Recommendation

Verify the immediate peer before trusting proxy headers; derive scopes from stored paired device records.

02 Fix review

Ship the fix. Apex re-tests it

Link a pull request and Apex replays the original proof of concept against it. Confirmed fixes get merged status; incomplete ones come back with what still reproduces.

Fix reviews

Apex re-tests each linked PR against the original proof of concept.

2 need attention
  • Fix confirmed OPEN-2

    Paired chat users can mint a setup code that silently auto-promotes a durable operator device

    OpenClaw audit · openclaw/openclaw · #287

    Fix confirmed in the latest review.

  • Fix in review OPEN-6

    First-time node connects expose system.run before node-pair admin approval

    OpenClaw audit · openclaw/openclaw · #296

    Fix review is checking the candidate patch.

  • Fix pending OPEN-3

    Spoofed Control UI locality silently mints persistent admin device tokens

    OpenClaw audit · openclaw/openclaw · #291

    Fix review still reproduces the vulnerable flow.

03 Scans & schedule

Connect in minutes. Findings in a few hours

Connect GitHub, pick repositories, and watch the scan progress live. Set a weekly or daily schedule and every new commit range gets the same treatment.

Audit scan running · scanning openclaw/openclaw 13m 57s
ETA ~15m · +3 new
First scan complete Schedule the next scan to keep this workspace current.
Schedule scans

Future scans start automatically on this cadence.

  • Weekly · Monday at 9:00 AM UTC (selected)
  • Daily · 9:00 AM UTC
  • Manual only

Next run: Mondays at 9:00 AM UTC.

04 Threat model

Hunts guided by your threat model

Before a scan starts, Apex writes a threat model for the repository: what an attacker can reach, which impacts matter most, and what to ignore. Edit it, then let the hunters optimize for it.

OpenClaw threat model

threat_model.md · Generated Apr 24, 2026

THREAT_MODEL

Scope Summary

This repo is an AI agent gateway/control-plane system. The highest-value findings turn untrusted chat, web, API, or plugin input into operator-equivalent control over the Gateway, paired nodes, browser state, secrets, or host execution.

Prioritized Impacts

  • Unauthorized host exec on gateway or node 100%
  • Sandbox escape to host or node 99%
  • Gateway auth / trusted-proxy / pairing bypass 98%
  • Credential theft with durable replay value 97%
05 Where you work

Findings land where your team already works

Summon a PR scan from a GitHub comment, or run the whole loop from your terminal: scan, watch progress, triage findings, and kick off fix reviews.

  • GitHub Comment @cantina-apex scan this pr and the review lands on the PR Check passed
  • Terminal apex scan, apex status, apex findings, apex findings fix-review Apex CLI
  • Trackers Send findings to Linear or Jira, or export Markdown for the audit report Export
  • Sharing Passphrase-protected, client-safe views of the workspace apex.cantina.security/share/8f3a…

Proof, not promises

Trained on real vulnerabilities, not synthetic data

  • 50,000+

    Real-world vulnerabilities analyzed

  • 9,000+

    Expert security researchers contributing signal

  • 10+

    Criticals and highs found in production code

  • $25B+

    In live funds secured

I was truly impressed by the subtle bugs that Cantina uncovered in an open-source cryptographic repository that I maintain, which had already gone through thorough reviews. Their AI-powered tool acts as a valuable safety net to catch bugs that humans and other tools may have missed.
Coinbase

Arash Afshar · Coinbase Cryptography Team

FAQ

Questions, answered

What teams usually want to know before connecting their first repository.

Apex uses AI-assisted code understanding to surface practical vulnerabilities, reduce false positives, and provide remediation guidance with context.
Yes. You can connect private repositories through your GitHub integration and control scan scope by repository, branch, pull request, or commit.
Yes. You can target specific branches, PRs, and commits, and include contextual documents to improve scan accuracy for your codebase.
Teams can comment on findings, validate or dismiss issues, and share views externally through passphrase-protected links.
Most teams can run their first scan in minutes after connecting repositories and setting up a workspace.