Architects Of Trust: HAQQ

For the longest time, I understood HAQQ through one meaning: justice. The Arabic word also carries the mean ing of truth. I understood the first when I founded HAQQ. Understanding the second took considerably more work.

Today, we serve more than 16,000 clients across over 200 countries and territories, with legal AI designed to work across laws, languages and jurisdictions worldwide. Our work extends from law firms to large enterprises and governments. I am proud of that reach. But the most con sequential thing we built emerged from a problem that initially seemed much narrower: how to build the archi tecture of trust into the fabric of artificial intelligence for law and other sectors where institutions cannot afford to trust AI blindly.

Law gave us an unforgiving education. A lawyer’s files contain other people’s freedom, fortunes, disputes and secrets. Confidentiality is part of the service itself. An answer must survive scrutiny, and a citation must lead somewhere real. A convincing explanation of the wrong law, in the wrong jurisdiction, is still wrong. I quickly under stood that making legal work faster would mean very little if we made it less dependable.

At first, I approached this as a data security problem. Where would information be stored? Who could access it? Could one client’s records surface in another client’s work? As we ad dressed those questions, I found a larger one underneath them. Even if an institution controlled where its data lived, did it control what happened when that data became intelligence? I came to believe that the future of AI would be shaped as much by who governs intelligence as by who produces it.

I began to understand sovereignty through that distinction. I wanted our clients to retain authority over the sources an AI system consulted, the rules it followed, the information it re membered and the actions it could take. Keeping a document confidential was one obligation. Preserving control over the de cisions built from it was another.

Meanwhile, the frontier models were becoming extraordinarily capable. ChatGPT and Claude forced a question I think every ap plication founder should confront: what would justify our exist ence as their capabilities improved? I could not build a durable company around the hope that they would stop getting better.

Nor could we realistically outspend the frontier laboratories to train a more powerful general model.

Yet raw capability left much of our problem unresolved. Hallu cinations, incomplete records, poor scans, outdated sources, latency and confidentiality all affected whether an answer was usable. I could give a model more intelligence and still give it the wrong evidence. Waiting for the next model release would not organize a client’s institutional memory or decide which em ployee was entitled to see it.

So we began building the machinery around intelligence. We developed ontologies to map relationships between docu ments, people, rules and jurisdictions. We vectorized databases so information could be retrieved by meaning. We built memory systems to preserve context and optical character recognition capabilities to make scanned records usable. We orchestrated agents to research, act and verify within defined permissions. Each component addressed something that stood between an impressive demonstration and work someone could actually rely on.

This changed how I thought about performance. A system equipped with the right evidence, context and controls could produce better results on a legal task than a more powerful model approached without them. We could improve reliability, reduce unsupported answers and lower the cost of completing work by controlling what needed to be retrieved, processed or sent to the frontier.

The pursuit of value had taken us from building an application to building the infrastructure beneath it. The application was becoming one expression of a much larger engine. Solving problems in law had led us into the architecture of sovereignty.

We brought that architecture together in Justinian, our sovereign AI platform. I think of it as an AI firewall: it sits between an institution, its knowledge and the models it uses. Its core processing and inference run locally, within the designated trust boundary. It governs access, retrieves evidence and determines how a task should be handled. Where external frontier capabilities are needed and policy permits, identifying details are replaced with stand-ins before external processing—a process known as pseudonymization.

That architecture gives us a reason to welcome every improvement at the frontier. We can draw on different models while keeping institutional memory, permissions and governance under local control. We can route work according to its complexity, cost and security require ments. The institution gains access to global intelligence while retaining authority over its use. It can change the model without surrendering what it has learned.
That is why I expect sovereignty to become more valu able as intelligence becomes more abundant. A model can be replaced; an institution’s knowledge, obligations and accountability endure. As AI moves from suggest ing decisions to executing them, I expect institutions to demand systems they can question, correct and govern. The ability to benefit from the next breakthrough while preserving that authority is, to me, a substantial part of AI’s future.

Our work with governments across the Middle East has strengthened that conviction. We are working on deployments at a scale I could scarcely have imagined when we began. Much of that work will become public in due course. For now, it has made clear to me how much is at stake: institutions want the benefits of AI while re taining responsibility for their own affairs. I believe our region has an opportunity to help define how that is done.

I keep returning to truth because sovereignty loses its purpose if the intelligence under our control cannot be examined. A local server can host a falsehood as eas ily as a foreign one. What matters to me is whether an institution can trace an answer to evidence, challenge its assumptions and establish who is responsible for acting on it. I see that ability to establish truth indepen dently as the ultimate guarantor of sovereignty.

I began with justice because that was the problem I understood. Following it far enough brought me to ques tions that belong equally to a hospital, a bank or a minis try. I now understand why our name was larger than the industry in which we started.

I founded HAQQ in pursuit of justice. That pursuit led me to truth, and to a larger conviction: the institutions that shape our world must remain sovereign over the intelli gence that shapes them, and sovereign from the labs that develop them. With that perspective in mind, law is only the beginning.