TCCC
The Construction Contracts Co.
Risk Scanner · Adjudicate Pro
← Back to the scanner

How we use AI responsibly

The Risk Scanner and Adjudicate Pro use AI to read your documents. Here is how that lines up with the Australian Government's Guidance for AI Adoption, the six essential practices from the National AI Centre.

Last reviewed: 28 September 2026. This guidance is voluntary and has no certification. We align with it; we don't claim to be certified.

Practice
What we do
1Decide who is accountable
One named owner: Mat Kennedy, who built the tools. He decides which AI models run, signs off every change to how they are used, and answers every complaint himself.
2Understand impacts and plan accordingly
The AI reads a contract and gives commercial information. It makes no decision about you: it doesn't approve, refuse or rate people. The main risk is a wrong or missed clause, so every report says it is general information, not legal advice, and you can challenge any report (below).
3Measure and manage risks
Instructions hidden in a contract don't steer the report: we tested contracts carrying hidden instructions aimed at the AI, and the report flagged and ignored them. Party names are kept out of reports. Daily spend and use limits switch the AI off automatically if something runs away. We keep a written incident plan.
4Share essential information
We say it is AI: at upload, on the report, in the PDF and in the email. We name the provider (Anthropic's Claude, via its commercial API), say where your data goes, and what the AI can get wrong.
5Test and monitor
Models are pinned to fixed versions and never upgrade themselves. Before any model or prompt change goes live, it runs on a fixed set of real contracts with automatic checks: report complete, the right state Act, no party named, every cited clause exists in the contract. Every live report gets that clause check too, and failures are logged for review.
6Maintain human control
You decide what to sign or send. The AI drafts, it never acts for you, and nothing goes to the builder unless you send it. Any scanner report can go to Mat for a human look with one tap. Mat can change the model or switch the AI off at any time.

Your contract and the AI

  • Anthropic's commercial terms say it may not train its models on your contract. We don't train AI models either.
  • The contract file is kept in Sydney for 12 months, then deleted automatically. Email us and it goes within 7 days. Full detail: Security and data handling.

What the AI can get wrong

  • It can misread a clause, miss one, or read a badly scanned page poorly.
  • It doesn't know what was agreed outside the document, or how a court would decide.
  • So the report is general information, not legal advice. For a big job, a deed, a guarantee or a live dispute, see a construction lawyer. See the Disclaimer.

Think the AI got it wrong?

Tap "Tell us" or the feedback buttons on your report, ask for Mat to look at it, or email mat@theconstructioncontracts.co. A person reads every one, and real errors feed the next round of testing.

The six practices evolve the earlier Voluntary AI Safety Standard (10 guardrails, 2024). Our mapping covers both, and we review it when the guidance or our system changes.