The precedent How it works Pricing FAQ
AI agent compliance & evidence

Your AI speaks for your company.
Prove it tells the truth.

ClaimSentry continuously verifies that your AI agents honor your contractual commitments — SLAs, warranties, pricing, terms — and generates timestamped, human-countersigned evidence: the due-diligence file for your AI.

See how it works
MCP grounding RFC 3161 evidence Available on-prem Patent pending · INPI FR2603347 · FR2603345
ClaimSentry · Agent audit — NordAssur
Commitments tested
✓
Home policy: 30-day cooling-off period
MATCH
✓
Claim response within 48h
MATCH
!
Water damage covered, no excess
OVER-ENG.
✗
Free cancellation anytime
CONTRA.
Agent under audit
NordAssur Assistant
OpenAI-compatible
Compliance score
Commitments honored 0%
2 honored · 2 flagged sealed · Evidence Pack ready
The precedent

Air Canada was held liable for
what its own chatbot told a customer.

In February 2024, a tribunal ruled that a company is responsible for the information its chatbot provides. The argument that the chatbot was a "separate legal entity" was rejected. What your agent asserts binds you.

Moffatt v. Air Canada
Civil Resolution Tribunal, British Columbia (Canada)
2024 BCCRT 149
Date February 2024
Holding Air Canada was held liable for inaccurate information given by its support chatbot. The airline's argument that the chatbot was a separate entity, responsible for its own actions, was rejected.

The reasoning is simple: a chatbot is part of the company's website, and the company is responsible for all of the information it provides — accurate or not. ClaimSentry lets you measure, correct and prove that your agents stay within your commitments.

How It Works

From your commitments to verifiable evidence

ClaimSentry turns your reference documents into a set of verified commitments, tests your agent against each one, and seals every verdict into an independently verifiable Evidence Pack.

1

Import your reference documents

Upload your terms, SLAs and warranties. ClaimSentry's LLM extracts the individual commitments; each one is reviewed and validated by a human before it becomes a test case.

2

Connect the agent to audit

Point ClaimSentry at the agent you want to test — your customer chatbot (any OpenAI-compatible endpoint) or a public AI. No integration into your product required.

3

Challenge & countersign

Each commitment is tested against the agent. A verdict is assigned on a six-level taxonomy — match, omission, over-engagement, under-engagement, wrong-perimeter, contradiction — then countersigned by a human with an audited justification.

4

Evidence Pack

You receive a report with RFC 3161 timestamps and independently verifiable .tsr files (checkable with openssl) — ready to drop into your compliance file.

Continuous monitoring

A model update can change your bot's answers overnight.

Schedule recurring runs, detect drift when the underlying model changes, and get alerted — with evidence — the moment an answer stops matching your commitments.

Be the first to know

Catch drift before your customers do.

ClaimSentry re-runs your commitment tests on a schedule. When a verdict shifts, a drift alert fires and the change is logged with full context and a fresh timestamp.

  • Scheduled runs — daily, weekly, or on your own cadence
  • Drift detection when the underlying model or prompt changes
  • Alerts via webhook, Slack or email, each backed by evidence
ClaimSentry · Drift timeline
Commitment monitored
"Free cancellation up to 24h before departure"
Run · Apr 02 MATCH
Run · May 07 MATCH
Run · Jun 11 · model updated CONTRA.
Drift alert sent · webhook + email
MCP grounding

Publish your verified commitments as a source of truth for your agents.

Expose the commitments you've validated as a machine-readable feed — an MCP server plus a JSON feed — so your own agents can ground their answers in your official facts.

For the agents you operate

One verified source,
for every agent you run.

The commitments validated in ClaimSentry are published via a Model Context Protocol server and a JSON feed. Your MCP-capable agents can query them in real time to stay aligned with your official commitments.

  • MCP server exposing your validated commitments, read-only
  • JSON feed for your own RAG pipelines and internal agents
  • You choose exactly which commitments are published

Scope: this grounds the agents you operate. It does not change how public assistants (ChatGPT, Gemini and others) answer — ClaimSentry does not promise to correct third-party AIs.

Read the MCP docs →
GET /commitments.json
{ "commitment": "48h claim response", "source": "CGV §7.2", "status": "verified", "verified_by": "human", "last_check": "2026-06-11", "evidence": "ts-1042.tsr" }
Sovereignty

Built for regulated industries.

On-premise deployment, one isolated database per instance, and an internal LLM judge — your data stays yours.

On-premise

Deploy entirely within your own infrastructure. Your reference documents and agent responses never leave your network.

Isolated by design

One dedicated database per instance. No shared storage, no cross-tenant data. Secrets encrypted at rest.

Internal judge

The LLM judge runs inside your perimeter. By default, no content is sent to any third party.

Compliance documentation

GDPR documentation and an AI-literacy note (EU AI Act, Article 4) provided to support your own compliance file.
Pricing

Three ways to work with ClaimSentry

Start with a one-off audit, move to continuous monitoring, or run everything on-premise. You keep control of your API keys.

Entry point
AI compliance audit · one-off
one-offOn request

A fixed-scope engagement. We audit your agent against your commitments and deliver a complete Evidence Pack.

Bring your own APIYou bring your own API keys — for the judge LLM and for the agent under test.
  • Commitment extraction, human-validated
  • Full challenge on the six-level taxonomy
  • Human-countersigned verdicts
  • Evidence Pack: report + RFC 3161 .tsr files
  • Fixed scope, fixed price
Available now
Continuous monitoring · SaaS
from290 €/mo

Fully managed in your own isolated instance. Scheduled runs keep watch on your agent and flag drift as it happens.

Bring your own APIYou bring your own API keys — judge LLM and agents under test.
  • Scheduled runs + drift alerts
  • MCP grounding feed
  • Email, Slack & webhook alerts
  • Quarterly Evidence Pack
  • Isolated instance, your API keys
Regulated sectors
On-premise · annual licence
from1 500 €/mo

Deployed in your own VPC or servers, billed annually. Your data never leaves your network.

Bring your own APIYou bring your own API keys — everything runs inside your network.
  • Everything in Continuous monitoring, self-hosted
  • One isolated database per instance
  • SSO · RBAC · full audit log
  • Internal LLM judge — no third-party calls by default
  • GDPR + AI Act documentation, onboarding included

Indicative starting prices — final pricing depends on scope (commitments, agents under test) and deployment. With your own API keys, LLM token costs are billed directly by your providers; ClaimSentry never marks up LLM usage. A Managed Judge option (calibrated judge included) is coming soon.

"We're running ClaimSentry with a first group of legal and compliance teams in regulated industries. Named case studies will appear here once these pilots conclude — we'd rather show real evidence than borrowed quotes."

CS
ClaimSentry pilot programme
Insurance · Banking · Telecom · Energy · Travel
FAQ

The questions legal teams ask

Is this admissible in court?
An Evidence Pack is not a court ruling, and we don't claim it is. What it gives you is a robust chain of proof: each verdict is hashed and carries an RFC 3161 electronic timestamp. Under Article 41 of the eIDAS Regulation (EU 910/2014), an electronic timestamp is admissible as evidence in legal proceedings and, when issued by a qualified timestamping authority, enjoys a presumption of the accuracy of its date and the integrity of the data. ClaimSentry supports qualified TSAs (dedicated configuration). Whether a given file is accepted, and what weight it carries, is for a court to decide — as with any evidence.
Does the EU AI Act require ClaimSentry?
No. No regulation requires ClaimSentry, and we never claim one does. What ClaimSentry gives you is the diligence and documentation file that supports your own obligations — for example AI literacy (AI Act, Article 4) and transparency toward users (Article 50, applicable from 2 August 2026). It is a tool to measure, document and prove; it is not a certification.
My chatbot isn't an OpenAI-compatible API.
Any endpoint that speaks the OpenAI-compatible format works out of the box. For other agents we provide adapters, and additional connectors are on the roadmap — tell us your setup and we'll confirm what's needed. The audit itself treats your agent as a black box: we send inputs and grade the responses, with no integration into your product.
Can you fix what ChatGPT says about us?
No — and we won't pretend otherwise. ClaimSentry does not change how public assistants (ChatGPT, Gemini and others) answer. What we do: measure what they say, prove it with timestamped records, alert you when it changes, and publish your reference commitments in a machine-readable format (MCP server + JSON feed) that your own agents can consume.