Strategic Legal Intelligence

Think about when the law usually meets a new technology.

A system is designed, purchased and switched on. It runs for months or years. Then something goes wrong — a person is wrongly denied a benefit, a business is wrongly penalised, a family is wrongly flagged. Only now do lawyers arrive. They ask the questions the law is built to ask: what happened, which rule applies, who is liable, what is the remedy?

Those questions are the right ones. But notice when they are being asked: after the harm, after the contracts are signed, after the institution has reorganised itself around the system, after thousands of decisions have already been executed. The lawyer is examining a building that is already occupied — and discovering that it was built without fire exits.

Every major failure of automated government has followed this shape. Australia’s Robodebt scheme ran for more than three years before it was stopped; the courts eventually declared its method unlawful, but by then hundreds of thousands of debts had been raised against people who did not owe them. Britain’s Post Office prosecuted more than nine hundred of its own sub-postmasters on evidence from faulty software; it took two decades and an Act of Parliament to undo the convictions. In both cases the legal system worked — eventually, expensively, and far too late to prevent the injury.

Strategic Legal Intelligence exists because “eventually” is not good enough.

Abstractions are easy to nod along to. So here is one concrete situation, followed through from beginning to end.

A provincial government is about to procure an automated eligibility system for a social-support programme. The vendor is foreign. The model is proprietary — the government will not own or fully see the logic. Rollout is planned within a year. Two hundred thousand households will be assessed by it.

On the day this procurement is approved, it looks like an IT project: a budget line, a tender, a delivery date. Nothing about it looks legal at all. That appearance is exactly the problem the method exists to correct.

Predictive — write down what is likely to go wrong, before it can

Prediction here is not prophecy. It is the discipline of looking at what has already happened when other governments deployed similar systems, and stating plainly what is likely to happen again: silent error at scale, harshest impact on the people with the least documentation, and an appeals process built for occasional human mistakes drowning under systematic machine ones.

The predictive work is one written document, delivered before the tender closes. Its value is timing: a risk written down before procurement is a design requirement. The same risk discovered after rollout is a scandal.

Preventive — put the safeguards where they are cheapest: in the contract

Once the risks are named, most of them can be prevented with ink, while the government still has all the leverage. The vendor must disclose error rates, measured on the province’s real population. The government — and any court that later reviews a decision — must have the right to audit the model: not a summary of it, the thing itself. A citizen must be able to see and correct the data used about them. The contract must fund a human-review channel sized for realistic volumes, so that “you may appeal” is a fact and not a decoration. And there must be an exit: data, decision records and continuity of service guaranteed if the relationship ends.

None of these clauses is exotic. What is rare is anyone with legal training being in the room, early enough, with the mandate to insist on them.

Protective — name what must survive the translation into code

Something subtle happens when a statute becomes software. The legislature wrote a purpose — support the poor. The department turned it into criteria. A contractor turned criteria into code. At each translation, small choices were made, and the accumulated drift can quietly invert the law: a support programme becomes, in operation, a suspicion machine.

The protective mode is a checklist of what must not be lost, stated before translation begins: the statutory purpose itself; equality before the criteria; privacy of applicant data; and above everything, the pair of abilities on which all other rights depend — a rejected applicant must be able to get reasons a human being can understand, and reach a forum with the power to actually change the outcome. A right of appeal to a body that cannot examine the algorithm is not a right; it is a ritual.

Proactive — build the institutions before go-live, not after the first scandal

Safeguards on paper need institutions behind them. What must exist, staffed and running, on the day the system switches on? Reviewers trained on how this system works and how it fails. A published explanation standard, so every rejection letter states what was decided, on what data, and how to challenge it. An incident protocol that answers, in advance, the question every institution eventually faces: if we discover systematic error, how do we pause the system, and who has the authority to press stop? Robodebt ran for years partly because stopping it was nobody’s clear job. And a duty to report to the legislature after twelve months — real numbers, error rates, appeal outcomes.

Prescriptive — put it all on paper, and give it to the people with authority

The final mode assembles everything into instruments decision-makers can act on: draft clauses, a review architecture, an oversight mandate, an implementation sequence, and criteria for adapting all of it as evidence accumulates. Two disciplines govern this step. Proportion: the prescription must match the evidence and the authority available. And humility: prescription is counsel, not command — offered to the people who hold lawful authority, to adopt, amend or reject.

What changed

Walk back through the story. Nothing about the technology was altered. The province may still buy the same system from the same vendor. But the procurement has changed its nature: it is no longer an IT purchase that happens to affect people, but what it always really was — a delegation of public power, now made lawful, explainable and reversible. The safeguards exist. The institutions are ready. The exit is real. And if something still goes wrong, the remedy is a working channel rather than a decade of litigation.

That transformation — from unexamined purchase to governed delegation — is what Strategic Legal Intelligence is for.

Everything above concerns a system already being procured. But the most consequential interval comes earlier: the period when a new form of power is emerging and the law has not yet comprehended it — when there are no cases, no categories, and no one is formally responsible for thinking about it.

That interval is when the most is still possible. It is also when dependency forms silently: a pilot becomes standard practice; a convenience becomes infrastructure; by the time anyone asks whether the arrangement is wise, unwinding it has become unthinkable. Anticipatory Legal Intelligence watches this interval deliberately.

Its cardinal discipline is honesty about uncertainty. Anticipation that presents one future as inevitable is not analysis; it is advocacy wearing analysis’s clothes. The purpose of looking ahead is to enlarge the options available to institutions, never to sell them a fate.