Find the key before someone bills you for it.
Paste a .env, repo file, CI log or MCP config. KEY/LEAK finds leaked AI and cloud keys, shows what each one can do, and tells you what to rotate first.
- Runs entirely in your browser
- Network blocked by page policy
- Nothing stored, values redacted
Trojanized LiteLLM 1.107.0 on PyPI and Docker
Reported guidance: pin to 1.106.2 or an earlier release, and rotate every secret that host could reach. Paste your requirements, lockfile, Dockerfile or compose file below and this page checks for the bad version. Free.
Reporting by SOCRadar and The Register. Details may change; check the vendor advisory.
Paste it. We'll read it.
Works on .env files, source files, CI logs, Docker and compose files, and MCP configs. Load several files at once if you like.
The sample uses fake keys generated in your browser. Anything real that you paste into any chatbot or website other than this one should be treated as exposed. Rotate it.
From panic to a plan in a minute.
Detect
Pattern packs for the AI and cloud providers that get abused most, plus a high-entropy check for named secrets in .env and MCP files. Placeholders and docs examples are ignored.
Rate
Every finding gets a damage rating and a plain-language blast radius: what the key can typically spend or reach if someone else holds it.
Rotate
A checklist per provider, ordered by damage, including what to look for in the logs after you revoke.
Pay once. No subscription.
Scanning is free. You pay when you want to know exactly what to do about it.
- Unlimited scans
- Exposure score
- Your single worst finding
- LiteLLM 1.107.0 incident check
- Blast radius per finding
- Rotation checklist
- Exportable report
- Everything in Free
- Every finding, worst first
- Blast radius for each one
- Rotation checklist ordered by damage
- Offline AWS account ID decode
- Copy or download a markdown report
- Everything in Unlock
- "Prepared for" client name on reports
- Print-ready branded report (save as PDF)
- For people who manage keys for clients
Payment is handled by Stripe. Your pasted text never touches it, because it never leaves this page. After paying you are redirected back here and the paid view unlocks on this device. Bookmark that confirmation link to unlock another device.
Leaked keys are the cheap way in.
- One leaked admin AWS key was enough to add an IAM user, subscribe to Bedrock models through Marketplace and run inference on the victim's bill (FortiGuard, Sept 2026, via CybersecurityNews).
- GitGuardian's 2026 report counts 28.65M new secrets on public GitHub in 2025, with AI-service secrets up 81% (report).
- Deleting a leaked key from a repo does not revoke it. One summary of the same report says 64% of secrets leaked in 2022 were still active in 2026 (summary).
- MCP config files are a growing leak spot: 24,008 unique secrets found in them on public GitHub, 2,117 still valid (Snyk write-up).
Straight answers.
Does anything I paste leave my browser?
No. The page ships with a content security policy that blocks all network requests, and the scanner is plain JavaScript running locally. You can verify it: load the page, switch to airplane mode, and scan.
Why not just use gitleaks or trufflehog?
They are excellent and free, and you should use them in CI. KEY/LEAK is for the moments they do not cover: you are on a phone or iPad, you have a pasted log or config, and you need to know what each key can do and what to rotate first.
Does it check whether a key is still live?
No, on purpose. Testing a key against a provider is itself risky and needs a network call. Check each provider's console, and rotate anyway.
What if it finds nothing?
That is not proof you are clean. Patterns miss custom formats and encoded secrets. Treat a clean result as one signal.
I already pasted a real key into an AI chatbot.
Rotate it. Treat anything you paste into any tool you do not control as exposed.
What do I get when I pay?
The full list of findings, a plain-language blast radius for each one, a rotation checklist ordered by damage, and a downloadable report. Agency adds a client name and a print-ready report.