We have signed an AI partnership with IBM Japan and I want to give some deeper insight on how we are working to maximise this partnership.
In short: we are embedding Arbitr directly across IBM’s primary platforms. This places our specialized agents inside their catalogue, our vertical models within their AI framework, and our operational workflows right in their developer suite, with their enterprise governance holding the verified audit log throughout.
We bring our own vertical small language models, our agents, the harness they run on, and the reviewers behind them. IBM brings the places enterprise customers already work: their agent catalogue, their AI platform, their developer tooling, and a governance standard those customers already audit against.
Our agents, in IBM’s catalogue. An agent is an application given one narrow job and allowed to do it end to end: check a term against a company’s approved glossary, test a claim against the rule that governs it, confirm a filing follows the required format. Ours run on the Arbitr harness. We are packaging them individually so that they can be deployed as independent agents inside watsonx Orchestrate, IBM’s catalogue of enterprise agents. The first one out is an AI Dubbing agent, released this month. Give it a video in one language and it returns the same video in another, in the speaker’s own voice, with the lip movement matched.
Our models, in IBM’s AI environment. What we build are vertical small language models. Small, because they are a fraction of the size of a frontier model, which makes them fast enough and cheap enough to run across everything a company publishes rather than a sample of it. Vertical, because each one is built for a single domain and a single job rather than for everything at once. We have built what we call a model factory, the production line that trains, tunes and deploys them, and it now deploys them into watsonx AI, so an IBM customer can run an Arbitr model inside the same environment they run everything else in.
Our platform, inside their developer tools. We are building gateways into IBM Bob, their AI coding tool. A developer working in that editor will be able to reach any model, agent or workflow built in Arbitr and integrate it into their own code, at the code level, without leaving what they are working in.
Their governance, holding our record. This is the one piece that runs the other way. Every time an agent acts, something has to record what ran, on what, against which rule, and what it decided, so a customer can prove afterwards how a decision was reached. We can use watsonx.governance to hold that record. What gets recorded is ours: the customer’s own rules, and the decisions their own people made. IBM provides the vault, not the contents. It is the least visible piece and the one regulated customers care about most, because it turns a system that works into a system they can defend to an auditor.
It might be helpful to share how we see the value on both sides.
The first is distribution. An agent in IBM’s catalogue is in front of IBM’s customers, and IBM’s sales teams can take it to them. A model that deploys into watsonx runs inside an environment those customers have already bought and approved. A gateway into their coding tool puts us in front of the developers who decide what gets built. None of that is distribution we could buy, and it reaches accounts we would otherwise spend a long time getting to. Every agent and model we add after the first one travels the same route.
The second is the standard. Large regulated organisations will not let a system loose on their own content unless it runs against controls they already recognise and already audit. Using IBM’s governance layer answers that objection once, rather than one customer at a time.
We are starting in Japan, and for a specific reason. Since April 2025, companies on the Tokyo Stock Exchange Prime Market have had to publish their earnings materials and timely disclosures in English at the same time as in Japanese. Most of them now do. The question is no longer whether they comply, but what compliance costs them, because the number of people qualified to do that work has not grown to match the volume and the deadline does not move.
We have completely rebuilt our SwiftBridge AI product with a new model trained specifically on Japanese financial data, agents that solve friction points we identified with testing we did across a number of prime listed Japanese customers and it is the first place this platform is being applied. Corporate disclosure is a good proving ground, because it is the setting where being wrong is least forgiving.
This is a build partnership. We expect the value to come through in new customers coming onto Arbitr, in how widely our agents are taken up inside watsonx Orchestrate, and in how our models get deployed and used. All of that sits inside the numbers we already report each quarter.
We have been an IBM build partner since 2023 and a Gold Partner since 2025. This is the point at which the relationship stops being a logo on each other’s slides. Our agents, our models and our workflows now run inside the tools IBM’s customers already use.
Thank you for your continued support.
David Sowerby
Co-Chief Executive Officer, Straker Limited