Responsible AI

1. What the engine does

Every clause in your contract is read and scored on three things:

Legal compliance — pass or fail against the Act that governs that contract type. There's no middle score here. A clause either meets the law or it doesn't.

Fairness — out of 100. A clause can be lawful and still be heavily one-sided.

Clarity — out of 100. A clause can be lawful, fair, and still unreadable by the person expected to sign it.

Clauses are also checked against each other, not only on their own. A salary and a set of hours can each look fine and together fall below the Award rate — that only appears when the contract is read as a whole.

2. Where the law comes from

Two different things get confused in this category, so we separate them.

The legal data is live. Award rates, penalty units, thresholds, commencement dates and the Acts that govern each contract type come from source, not from what a model absorbed during training. When a rate changes on 1 July, the engine works from the new one. When an Act is passed but never commenced — as with Tasmania's 2022 retail leases legislation — it isn't treated as law, because it isn't.

The analysis is a model, and models have limits. Reading a clause and judging it is not the same as looking up a rate. Section 5 covers where that goes wrong.

Contractam is built for Australian law only. It does not analyse contracts governed by the law of another country.

3. What it checks in its own work

Before an analysis or a generated contract reaches you, the engine reviews its own output and re-runs the work where the result doesn't hold up — up to three times.

This is not a human reading your contract. It's the system checking itself against the same legal sources it used the first time. It catches errors. It does not catch all of them.

4. What it won't do

This is the part most AI tools leave out.

It won't answer questions that turn on facts it can't see. A contract shows what was written, not what was said in the interview, what's been happening on site, or how the relationship has actually run. Where the answer depends on any of that, Louis says so rather than guessing:

"That one needs a lawyer. It turns on facts I can't see from the document."

It won't tell you what to do. It will tell you what the law requires, what a clause commits you to, and where a contract falls short. Whether to sign, push back, escalate or walk away is your decision, and the engine is not equipped to make it for you. Our Disclaimer sets out where the service stops and when to get a lawyer.

It won't invent a clause that isn't there. Where your contract is silent on something, it says so instead of filling the gap.

It won't guarantee an outcome. A rewritten restraint is drafted to be defensible, never described as enforceable. No engine can promise how a court will read a clause, and any tool that does is telling you something it can't know.

A system that never refuses isn't being careful with you. The refusals are deliberate.

5. Where it gets things wrong

Artificial intelligence is probabilistic. Ours is capable and it is not infallible.

It can misread an unusual clause. It can miss something in a long or heavily amended document. It can flag a standard term as a problem, or treat something significant as routine. It can produce a confident answer that is wrong — that's the failure mode worth knowing about, because nothing in the output will look different when it happens.

Check clause references against your original document. If the analysis points to clause 8.2, it's worth two seconds confirming clause 8.2 says what the analysis says it does. That single habit catches the errors that matter most.

6. What people do, and don't do

No lawyer at Contractam reviews your contract. No person reads your document before the analysis reaches you. The engine runs on its own.

People at Contractam build and test the engine, review its performance in aggregate, and investigate errors you report. They access individual documents only when you ask for support on one, during a security investigation, or where the law requires it.

7. Your documents are not training data

We do not use your contracts, or the contracts we generate for you, to train any AI model. Not ours, and not anyone else's — our agreements with our AI technology providers prohibit them from training on your documents too.

This isn't limited to public models. It's absolute: your documents are processed to produce your analysis, and for nothing else. How your documents are stored, who can reach them and how long we keep them is set out in our Privacy Policy.

8. Telling us it got something wrong

If the engine gets something wrong, we want it. Flag it in the analysis itself, or email [email protected] with the contract and the clause.

Reports go to the team that maintains the legal sources and the scoring framework. Where an error comes from stale or incorrect legal data, we correct the source — which fixes it for everyone, not just for you. Where it comes from how the engine reasoned, it goes into the testing set we check releases against.

Correcting our legal sources and our tests is not the same as training a model on your document. Your contract stays out of it.