Fail-closed
When Verity can't verify a claim, it says so. It never calls a guess the truth — silence beats a confident lie.
The trust layer for AI
Your AI is making things up. We catch it first.
Verity is an independent, fail-closed verification engine. Paste an answer your AI gave a user — it extracts every factual claim, checks each one against the live web, and flags the fabricated ones with the evidence and a calibrated confidence score. Before your users do.
Scan your AI — free
For agents & devs — from $0.02/call
No signup · free scanner · pay-per-call API from $0.02
AI-generated draft · submitted for review
In support of plaintiff's motion, counsel relies on Henderson v. Allied Mutual, 412 F.3d 887 (9th Cir. 2005), which the brief asserts established a duty of continuous disclosure. The argument is further grounded in Carrington Trust v. Delaney, 559 U.S. 214 (2010), holding that silence may constitute misrepresentation. By contrast, the controlling authority on a passenger's federal remedy remains Zicherman v. Korean Air Lines Co., 516 U.S. 217 (1996).
Findings
Henderson v. Allied MutualFABRICATED
No matching reporter entry. No 412 F.3d 887 case by this name in any federal database.
Carrington Trust v. DelaneyFABRICATED
Citation 559 U.S. 214 resolves to an unrelated 2010 opinion. Party names not found.
Zicherman v. Korean Air Lines Co.VERIFIED
516 U.S. 217 (1996) confirmed against the primary source. Genuine — left un-flagged.
Signed verdict
2 fabrications caught in this sample · 1 genuine citation cleared
sig: ed25519 · 9f3c…a17e · fail-closed
fabricated citations caught
0 false positives on the genuine cases the Mata v. Avianca benchmarkThe proof
We ran Verity against the citations that sanctioned real lawyers.
In 2023, two U.S. attorneys were sanctioned in the U.S. District Court for the Southern District of New York after filing a brief built on fake cases their AI invented — Mata v. Avianca. We fed Verity those exact fabricated citations.
On the Mata v. Avianca benchmark, it flagged every fabricated citation — 8 of 8 — with zero false positives on the genuine cases. This is the only hard accuracy number we publish, because it's the only one we've verified.
Why Verity
Three promises
most guardrails can't make.
Everyone says they catch hallucinations. The difference is what happens when they're unsure — and whether they have any reason to tell you the truth.
Calibrated
Every finding carries an honest 0–1 confidence score and the evidence behind it. You see exactly how sure we are, and why.
Independent
We don't ship the AI we grade — so we've no reason to grade yours soft. Just the truth about what's wrong.
How it works
Paste an answer.
Get the truth about it.
Submit
Paste any answer your AI gave a user. No setup, no signup on the free scanner.
Extract
Verity pulls out every individual factual claim — names, dates, citations, numbers.
Verify
Each claim is checked independently against the live web, with the evidence captured.
Verdict
Fabrications flagged in red with a calibrated score. Genuine claims left untouched.
Every other guardrail is one cloud's API or four separate bills. Verity is one fail-closed, signed verdict.
The full audit
A reliability report
you can hand to a regulator.
What you get
One fixed-price pass over everything your AI is shipping.
For teams shipping into legal, health, and fintech: a liability-ranked report of every hallucination, injection risk, and PII leak — plus a public reliability page you can show customers and regulators.
- A liability-ranked report of every hallucination, injection risk, and PII leak
- A public reliability page you can show customers and regulators
- Delivered in 5 business days — independent and fail-closed throughout
For developers & agents
The same verdict,
keyless and pay-per-call.
The exact verification engine, exposed as a pay-per-call API on the x402 standard over Base mainnet. No accounts. No API keys.
Your agent POSTs to an endpoint, gets a 402 with the price, pays USDC automatically, and receives a calibrated, fail-closed JSON verdict. From $0.02 / call.
No keys · no accounts · USDC per call · MCP server in the official registryAgent POSTs
Your agent POSTs the claim to a Verity endpoint — no key, no signup.
POST /verifyGets a 402
The endpoint returns a 402 with the price for this call — settled on Base mainnet.
402 Payment RequiredPays USDC, gasless
The agent pays USDC automatically and gasless — from $0.02 a call. No invoice, no contract.
pay · USDC · BaseCalibrated verdict
Receives a calibrated, fail-closed JSON verdict — claims, confidence, and evidence.
200 · verdict.jsonOne engine, a whole verification suite.
Each endpoint is independent, fail-closed, and pay-per-call. Prices are per call.
Verity/verify
Extract claims and verify each against the live web.
from $0.02 / callGuardrail/check
Screen an answer for policy and safety violations.
from $0.02 / callGround/ground
Ground a response in cited, retrievable sources.
from $0.03 / callVerify-ID/source
Confirm a cited source actually exists and matches.
from $0.02 / callCitation/cite
Validate citations against the genuine record.
from $0.03 / callProvenance/provenance
Trace where a claim or asset originated.
from $0.02 / callSentinel/sentinel
Detect prompt-injection and manipulation attempts.
from $0.02 / callRedact/redact
Find and redact PII before it leaks to a user.
from $0.02 / call
Plus Arbiter · Freshness · Sieve · Compass · Concord · Tally · Distill · Faithful
— see the full catalog and live prices.
Open catalog
Start now
See what your AI
made up today.
The audit is a fixed $750, delivered in 5 business days: a liability-ranked report of every hallucination, injection risk, and PII leak — plus a public reliability page you can show customers and regulators.