tokenizen.nl · Eindhoven

A payment proves money moved.
We make what arrived checkable.

AI agents now pay each other. Our open-source receipts record what the seller offered and what the buyer says arrived, each signed by its own side, so anyone can check them. We also find leaks in real software and show the numbers. In an open-source AI tool with more than 10,000 stars, 33 secrets leaked before, 20 after the maintainers' own fixes and 4 with ours (pull request open). Check your repository free in a minute, see the price of a smart contract scan before you pay, and keep a log whose timestamp is outside our control. Every number below links to the commands, so you can run it yourself.

• 0 of 600 live x402 resources in our October sample sign a receipt. Ours does, and you can verify it
• On-chain one real AI-agent payment on Base mainnet, with the commands to check it yourself
• 109 public repositories in ten EU countries measured on eight security signals, our misses published
• Recount our scan results come with every verdict and a script that recounts them, the leader's higher score included. Someone outside our team has already recounted them

tokenizen

What we do.

Six things. Each has its own page with what you get, how you check it and what it is not.

Reliability

Our numbers say where to check them.

• Each result lists the public pages that were asked for, and whether that worked.
• The timestamp on a Signal log can be checked without us.
• Our misses sit on the same page as our hits.

Technique

We measure first and claim after.

• Eight security signals, measured over 109 public repositories on 28 October 2026. The method is in a published note.
• A log in which every line holds the hash of the line before it, with a timestamp from OpenTimestamps.
• Open-source tools on GitHub and npm: capacity-attest, agent-receipt-verify, x402-lint and muraqib-agent-watch.

What sets us apart

Dates you can check without us.

• The timestamp of a Signal log comes from OpenTimestamps, which is outside our control, and shows that the log existed by that date.
• We publish what our scan missed: 97 of 105 known findings in its run on eight audit contests.

Security

Security research. What we can show, and what we cannot yet.

Some of what we found we may show, some we may not, and some will follow when the fix is out.

• public

  A published GitHub security advisory, severity High, with our account credited as reporter.

  Open the advisory

• our own records

  Reports confirmed or assessed by two known technology companies. At two more, the report was received or fixed without an answer to us. The names stay private.

• planned

  Planned, not live yet: each report we file would get a date and a fingerprint on a public list, without its content. When the fix is out we would publish the dossier, and anyone could see that we had it on that date.

The whole record

Mission and vision

Mission

Make security claims that anyone can check.

We build checks, publish how they were made and publish where they fail.

Vision

A web where every claim arrives with the way to check it.

Software, agents and payments that prove what they say, so that trust is something you measure.

Free

Now check yours.

Type owner/name of a public GitHub repository and see which of the eight signals are visible in it.

We read public pages only and store no report.

What we do not know, and where we missed.

The numbers that do not flatter us sit on the same page as the ones that do.

• Our smart contract scan found 8 of 105 known findings in its first 15 hypotheses, on contests that were no longer secret, after two changes at once. It missed 97. In the sealed run the earlier versions found 1 and 4.
• In 29 of the 109 repositories we could not tell whether a lockfile is present, so they count neither way.
• The email warnings of the Signal log are not built yet.

Scoreboard with every run

Conversation

Talk to us.

Start with a conversation about where you stand. Ask for our measurements and we show how each one was made.