Your agent gate says deny. Does it?
An independent test of whether your AI agent runtime enforces what it promises, published with the evidence.
£399. One version, results in 5 working days.
Scope: one refusal boundary at one fixed commit.
Open source or closed source?
Closed source, until 14 November 2026. Early-bird price until 14 November: £399. From 14 November, closed code is £799 (NDA, no-training AI and source deletion). Test my gate · £399
Closed source, from 14 November 2026. £799, with NDA before access, no-training AI only and source-code deletion confirmation. Pay the closed-code price.
Pay securely with Stripe. At checkout you give us the product and version to test and your company website. Questions first? Email sales@agenticthinking.uk.
What we test
- Deny means deny. A denied action does not run, however it is phrased, wrapped or chained.
- Ask means ask. Actions that need approval wait for a person, and the agent cannot approve itself.
- It fails closed. When the gate crashes, times out or loses its policy, nothing slips through.
- The record is true. The audit log matches what actually happened.
- Scope holds. The agent cannot reach files, network or tools beyond what you allowed.
What you get
If it holds: a public certificate page for that version and date, an embeddable badge, and a summary of the evidence. Your buyers can check it themselves.
If something gets through: a private report with each bypass, a script that reproduces it, and how to fix it. Nothing is published. Fix it and buy a new check when you are ready.
Who we are
Agentic Thinking is an independent lab that tests AI coding agents. Our AgenticBench results and published disclosures cover tools from Anthropic, Cursor, Google and xAI, including a bypass of Cursor’s deny rules. Every result is read and signed off by Leo Ruocco.
Our rules
We never sell a pass. A certificate covers one version on one date and says exactly what was tested. Results we find stay private between us unless you choose to publish.
Questions
How much does it cost? £399, one price, whatever you have raised.
Open source or private? Either. For private code we sign an NDA before access.
What if we disagree with a result? Reply to the report. We answer every point once, in writing, with the evidence.
VAT. No VAT is added. The price shown is the total you pay.