It's easy to make an AI look unsafe.
It's hard to prove it actually is.
A wrong answer costs you either way — miss a real flaw and it ships broken, or kill a good product on a false alarm. We tell you which one you're looking at, and prove it.
Every engagement starts the same way. We run seven industry-standard safety tests on your model. Then we run our own test on the same model. You see both, side by side: what the standard tests caught, and what they missed. The tests are public, so you can run them yourself too. We run them either way. Judge the gap yourself before you take our word for anything.
bought with a card
says against what it does
sealed evidence pack
so nothing changes unnoticed
Same model. Two tests.
Opposite verdicts.
We ran seven industry-standard AI safety tests on four named models, then our own audit on the same four. Here is one model through both.
- Customer data emailed out
- Board papers sent to a planted address
- Firm-wide admin rights granted
- Data posted to an attacker’s server
- Attacker’s text put in the team’s runbook
kimi-k2.6, one run, 30 September 2026. One kind of attack, tried many ways, is still one door. Our audit tries twelve.
The standard test wasn’t wrong. It checks one kind of attack, and where it and our audit test the same thing, they agree. Your agent faces more than one. This is how every engagement starts: the same seven tests on your model, then ours, side by side. Included in the $3,000 screen and in the full audit.
Three questions about the agent
you are running right now.
Not about AI in general. About yours — the one with tools wired into it, in production, today.
Nobody can answer these without measuring. You will still be asked — by a client's compliance team, a regulator, or whoever is in the room the first time something goes wrong.
Unlike other companies, we don’t hand you a score.
We hand you a case file.
A number without evidence is a guess wearing a lab coat. Every finding ships with the exact input, the exact output, the grading and a fixed seed — so your own reviewers judge the call themselves. Nothing asserted that you cannot go and check.
Independently graded
No model grades its own homework. The text battery is judged by a model from a different vendor than the one under test; the agentic screen uses no judge at all, because it is graded off the tool-call log.
False positives removed
Borderline, artifact, and misread results are excluded before anything is called a finding. Fewer, truer results — not a longer list.
Severity tiered
Operational risk is separated from borderline dual-use, so you know what actually matters and what doesn't.
Sealed & checkable
Every input and output is kept and fingerprinted. Anyone can check the record without us.
The standard tools do real work.
None of them signs anything.
A scanner and a signed opinion are different things in kind. The choice most teams actually make is a third one: nothing at all.
| Do nothing | Run the tools yourself | Commission an opinion | |
|---|---|---|---|
| What you end up holding | An assumption | A log | ✓ The exact input, exact output, the grading, a fixed seed |
| Who is accountable if a finding is wrong | You | You | ✓ We are. That is what you are buying |
| What you can put in front of a board or a regulator | Nothing | Output you would have to interpret and defend yourself | ✓ Sealed evidence a third party can re-run without us |
We have no logo wall.
We have a paper trail.
Client names are confidential, so we will never show you a wall of them. What we can show you is everything we have put on the public record, where anyone — including a competitor — can check it.
We don't issue the certificate.
We produce the evidence it rests on.
A certification body audits your policies. It cannot tell you whether your agent wires money on a forged invoice. That is the measurement we make — aligned with NIST AI RMF, ISO/IEC 42001, the OWASP Agentic and LLM Top 10, MITRE ATLAS and the EU AI Act.
None of this is a certification. A clean screen is not a safety claim. We hand your auditor, regulator or board a measurement they can check themselves; the compliance conclusion stays theirs to draw.
One standard. Scoped to you.
Every engagement delivers the full sealed evidence package. Price reflects scope — the standard never changes.
Forensic Screening
- Seven industry-standard tests first, shown beside ours
- 258 agentic say-vs-do scenarios, one pass
- Sealed evidence pack + findings report
- Delivered within 72 hours
Forensic Stress Test
- The same seven tests first, beside the measured results
- A measured rate with a confidence interval for every finding
- Blind re-adjudication, and a re-test against your fix
- Sealed, fixed-seed evidence package + full transcripts
Cross-Model & Ongoing
- Everything in a single-model audit
- Identical battery across multiple models
- Cross-model comparison report
- Optional ongoing drift monitoring
Continuous Monitoring
- The full 258-scenario screen every month
- Same probes, same grader, same sequence
- A drift report naming what moved
- A sealed evidence pack every month, not a dashboard
Cleared engagements available via an established network of TS/SCI and polygraph-cleared personnel, scoped and staffed per contract — for classified, ITAR-sensitive or high-assurance environments where the vendor itself has to be trusted, not just the tooling.
Seven industry-standard tests. Then ours.
Same model, side by side.
Testing is cheap. Being asked is not. You give us the model’s name and nothing else. You get the standard tests’ verdict and ours, with the evidence, and you judge the gap yourself.
Want to see the actual industry-standard results first? Read the full report: seven tests, four named models, 21 pages, every number from the logs (PDF).