Inoku is accepting private beta testers

Larnaka, Cyprus22 October 2026

Inoku, an independent research lab working on whether AI agents tell the truth about their own work, today opened applications for the private beta of its evaluation tools. The beta is free. The first wave of invites goes out in Q1 2027, later waves follow, and every applicant gets a reply with their wave.

The problem the tools measure is documented. In one 2026 study, false success, an agent reporting a task complete when it was not, made up 75.8% of failures among coding agents that reported their own status, and the best LLM judge reading the transcript caught it at an AUROC of 0.65 (Advani, 2026).

What beta testers get

• The first builds of the tools behind Inoku Verified™: run an agent on a set of tasks, keep an independent record of what it did, and report how often its “done” and “I couldn’t” were true.
• Early images of Inoku OS, the Linux distribution those tools will ship in. It is in development, and testers should expect rough edges.
• The results of every run they make, theirs to keep, and a direct line to the people building the tools.

Who it is for

Teams that build or deploy coding, operations or data agents on real systems, and can run the tools on a Linux machine or virtual machine they control. Testers run the tools on their own tasks and report what breaks, what confuses them and what is missing.

What the beta is not

Nothing in the beta is for sale, now or when it ends. The beta measures; the Inoku Verified™ mark comes later, under rules Inoku will publish before the first mark is granted. The method itself is described when it is ready to be checked, not before.

Quote

“An agent’s ‘done’ is a claim, not a fact. We want to measure how often it is true, on real tasks, with the agent kept out of the record. We also measure the opposite: how often an honest ‘I couldn’t’ was right, because an agent that stops when it should is worth more than one that always says done. The beta is how we find out what breaks before we publish,” said Ivan Popov, founder of Inoku.

How to apply

Applications are open at inoku.com/beta. Teams that want Inoku to run evaluations for them, rather than run the tools themselves, can apply to the design-partner programme instead, or as well.

About Inoku

Inoku is an independent research lab based in Larnaka, Cyprus, founded in 2026 by Ivan Popov. It builds ways to measure and reduce false success claims by AI agents: cases where an agent reports a task complete when it is not. Its work covers an independent evaluation and mark (Inoku Verified™), private evaluations for model builders, task environments with an independent ground truth, and an independent record of what production agents did. Inoku plans to publish its tools and benchmark under an open licence. Every factual claim on inoku.com cites its source.

Media contact

Ivan Popov, hello@inoku.com. Logo files and usage rules are on the brand page. Inoku on LinkedIn: linkedin.com/company/inoku-labs.