SignetAI
Lagos · Abuja · Alberta
§ Notes · short pieces from the Signet AI team

Notes — what we’ve learned building AI systems institutions can trust.

Short, plain pieces on the things that actually matter when you put AI into a serious organisation. We publish when we have something useful to say — not on a schedule.

N° 01 MMXXVI · April By the Signet AI team ~5 min read

Why it matters that the AI shows its sources.

It’s easy to build an AI that sounds confident. The hard part — and the part that counts — is one you can actually check.

Plenty of AI projects launch without answering the one question that decides whether a serious business can use them at all: when the AI gives you an answer, can you tell where it got it? That’s the question that matters — and it has almost nothing to do with how clever the AI sounds.

Read on →
trusthow it works
N° 02 MMXXVI · March By the Signet AI team ~4 min read

Why we keep a record nobody can change.

Most software keeps logs so engineers can fix bugs. A serious business needs something stronger: proof.

The first record-keeping system we built for a client was an ordinary database table. It worked fine. Then their compliance officer asked one question that changed how we build everything since: “Can you delete a row?” Technically, yes. The answer she needed was no — and meant it.

Read on →
trustsafety
N° 03 MMXXVI · February By the Signet AI team ~4 min read

Knowing how the AI can go wrong is half the job.

If you can’t describe the ways your AI will make mistakes, you can’t stop them. Most AI ships without anyone doing that work.

Our system can refuse to do something in about eleven different ways, and every one of them comes from a real mistake we’ve watched an AI make: the right email to the wrong person, a figure pulled from an out-of-date document, a letter written in a tone the company would never use. Each mistake became a named check the system now runs before it acts.

Read on →
safetydesign
N° 04 MMXXVI · January By the Signet AI team ~6 min read

Why your data should stay in your country.

The cheapest place to run your AI is rarely the right one. For most African organisations, the right answer is at home.

A government agency we work with was once offered an AI trial at almost no cost — as long as their documents could be processed on another continent. It would have worked. It would also have handed away control they could never easily get back. They said no. They were right.

Read on →
your datastaying in control
More to come

The next pieces in the queue.

What “a person always approves” means when nobody wants to be the one who signs. Why a dashboard isn’t the answer. What we learned from the first hundred times the AI said no.