The Independence Manifesto
Why the one thing verification has to be is the one thing disappearing from it — and what we're doing about it.
Verification has a simple job: tell you whether a claim is true, from a vantage point that isn't the claim's author. The whole value is in the second half of that sentence. A grade means nothing if the thing being graded also holds the pen. And yet, across the AI-trust market in 2026, the independent grader is quietly going extinct — bought, absorbed, or replaced by pipelines that check their own homework.
The independent slot is vacating
Look at who was selling independent AI trust and verification eighteen months ago, and where they are now.
- Lakera was acquired by Check Point (announced September 2025; terms undisclosed, media reports put it near $300M).1
- Galileo was acquired by Cisco (announced April 2026, completed May 2026).2
- Protect AI was acquired by Palo Alto Networks (reported at over $500M).3
- Prompt Security was acquired by SentinelOne (a deal reported at approximately $250M).4
- Patronus AI — the one most directly in the runtime-evaluation slot — pivoted away from it, raising $50M to build simulation environments that stress-test agents instead.5
- kluster.ai sunset its verification services on June 9, 2026.6
Every one of these was, at some point, a company you could pay to independently check an AI's output. Today, most of them are a feature inside a larger security platform — and platforms have their own products to protect. The remaining path many builders reach for is worse for exactly the reason acquisition is: you let the vendor grade its own pipeline. Vectara's hallucination scoring evaluates Vectara. AWS's Automated Reasoning and Bedrock Guardrails check policies you author on AWS's own stack. There's nothing wrong with a platform offering built-in guardrails — but a guardrail written and scored inside the same system it's guarding is not independent verification. It's the system vouching for itself.
So here is the contradiction at the center of the market: the single property that makes verification worth anything — independence — is precisely the property being consolidated out of existence. The more the category succeeds commercially, the less independent it becomes.
What we refuse to be
VerityLayer exists to fill that vacating slot and stay in it. We are an independent notary for AI claims: you (or your agent) send a claim, we return a verdict, the evidence behind it, a calibrated confidence, and — when the evidence doesn't support a confident answer — an honest abstention rather than a confident guess. We don't sell the model being checked. We don't own the pipeline being graded. That's the entire point.
But independence is easy to claim and hard to prove. Any vendor can put "independent" on a landing page. Verification you can't check yourself is just another vendor asking for blind trust — and we refuse to be that. Blind trust is the exact problem we're supposed to solve; smuggling it back in through our own marketing would be a self-inflicted wound.
Receipts turn "trust us" into "check it yourself"
So we did the thing that makes the claim checkable. Every verdict VerityLayer returns is now cryptographically signed — an Ed25519 signature over the verdict, the claim hash, the timestamp, and the model tier. We publish our public key at a fixed, open location, and we run a free endpoint where anyone can re-verify a receipt independently — no account, no payment, no permission from us.
As far as we can find, no other AI verification service signs its verdicts. We're not going to call that a "world first" — that's exactly the kind of unprovable claim this whole essay is arguing against. What we'll say is only what we checked: we looked, and we couldn't find another one. If you find one, tell us, and we'll say so.
A signed receipt changes the relationship. Today, verification asks you to take its word. A receipt lets you take the math's word instead. You can hand a VerityLayer verdict to a third party — a court, an auditor, a customer, a skeptic — and they can confirm it came from us, unaltered, without asking us anything. That's what independence looks like when it's real instead of rhetorical: not a promise, an artifact.
Verify it yourself
We'll close the way the whole thesis demands — by not asking you to believe us. Here is a live, signed, self-verifying receipt you can check right now, for free, from your own terminal:
curl https://api.veritylayer.dev/receipt/selftest
Our public key lives at api.veritylayer.dev/.well-known/verity-pubkey.json, and you can re-verify any receipt yourself with a free POST to api.veritylayer.dev/receipt/verify.
We're early and small — a handful of settled payments from one outside wallet, on the order of a dollar. We're not going to dress that up as revenue. But the principle is fully built and running today: the independent notary that stays independent, and proves it with every receipt.
Because verification you can't check isn't verification. It's just one more thing you're being asked to trust.
Sources
- Check Point press release, 2025-09-16: "Check Point Acquires Lakera to Deliver End-to-End AI Security for Enterprises." checkpoint.com (terms undisclosed; ~$300M per media reports). ↩
- Cisco Newsroom, 2026-04-09: "Making AI Trustworthy and Observable in Real-Time: Cisco Announces Intent to Acquire Galileo." blogs.cisco.com (completed May 2026). ↩
- Palo Alto Networks press release, 2025-04-28: "Palo Alto Networks Announces Intent to Acquire Protect AI." paloaltonetworks.com (completed July 2025; price reported >$500M). ↩
- SentinelOne press release, 2025-08-05: "SentinelOne to Acquire Prompt Security to Advance GenAI Security and Agent Security Strategy." sentinelone.com (~$250M reported; completed September 2025). ↩
- TechCrunch, 2026-06-25: "Patronus AI lands $50M to build digital worlds that stress-test AI agents." techcrunch.com ↩
- kluster.ai verification services sunset June 9, 2026; "kluster.ai has joined MITO" per its own homepage. kluster.ai ↩