Frequently asked questions

Do we need an AI policy before we deploy AI tools?

Yes, and the expectation now sits ahead of the law. Audit offices have called for governance to be in place before deployment, and a national framework is taking shape: an Office of AI was established within the Department of the Prime Minister and Cabinet on 15 July 2026, Australian Standards for AI go to National Cabinet in August 2026, and legislation is expected in Parliament in early 2027. Procurement clauses and QA expectations arrive before any of it. The practical test is whether you could show today which AI tools your organisation uses, who approved them, and how their output is checked before it reaches a client or a regulator.

What does the NSW Audit Office expect of councils?

Fit-for-purpose governance arrangements in place before AI is deployed, and a readiness review that fell due on 30 June 2026. Only 40% of NSW councils have established a formal AI policy or embedded consideration of AI into existing policies, while 90 councils are already running 109 AI tools between them, so most of the sector is deploying ahead of its own governance (Audit Office of New South Wales, Local government 2025, January 2026). No mandatory framework yet applies to councils; the Audit Office has recommended that the Department of Planning, Housing and Infrastructure set one, which leaves each council establishing its own arrangements in the meantime.

We already have an AI policy. What else is there?

A policy states intent. What auditors, insurers and procurement teams ask for is the layer underneath it: a register of the tools actually in use, validation and QA standards for AI-assisted work, a disclosure practice, and evidence that a qualified person reviewed the output before it left the building. The first of those is the usual gap. Only 10% of NSW councils could account for every AI tool they had implemented in a central inventory, and only 11% held an AI adoption strategy (Audit Office of New South Wales, Local government 2025, January 2026).

Can AI write our environmental reports?

Parts of them, under supervision. Data analysis, monitoring and trend reporting, document review and first-draft text are all faster with AI, and those gains are available today. Interpretation, professional judgement and sign-off stay with the practitioner.

One hard limit applies before any of that. In proceedings before the Land and Environment Court of New South Wales, generative AI must not be used to draft or prepare the content of an expert report, or any part of one, without prior leave of the Court.

The rule is narrower than its reputation. Spelling and grammar correction, formatting, transcription and translation sit outside the practice note altogether, and chronologies, indexes and the summarising and review of documents are expressly permitted. Leave is available for the rest, and the Court’s own examples of leave-worthy use are interrogating very large data sets and conducting statistical analysis, which is the kind of analysis environmental evidence often rests on. An application names the tool and version, the benefit sought and the documents to be submitted to it; where leave is granted, the expert discloses which parts of the report relied on it and annexes a record of how it was used (Land and Environment Court of New South Wales, Practice Note: Use of Generative Artificial Intelligence (Gen AI), commenced 12 February 2025). Comparable restrictions apply in other NSW courts.

Any report that could become expert evidence therefore sits outside the drafting use case altogether. Knowing which of your work that covers, and where the permitted uses stop, is a procedural job.

Will AI-assisted work survive an audit or a regulator review?

It survives when you can show how it was checked. That means a record of which tool was used, what source data each figure traces back to, who reviewed the output and against what standard, and a disclosure practice that matches both your client agreements and the court requirements above. Without that record you are defending the work on assertion alone.

Is our data safe if we use these tools?

That depends on which tool, on which plan, and on what actually leaves your network. The distinction that matters most is between sending a dataset to a hosted model and having the model write the analysis while the code runs on your own machine, against data that never goes anywhere. Data analysis, statistical work and trend reporting all fit the second pattern, and it is the one we design for wherever the data is sensitive. Where material does have to go out, three controls do most of the work: a short list of approved tools, an agreed step that strips identifying detail from the prompt, and a service tier that contractually excludes your inputs from model training.

How is this different from a general AI consultant?

General AI consultants rarely know what an EPA notice, a site audit or a contaminated-land assessment means in practice. On the other side of the gap, few environmental consultants run AI deeper than a chat window. Grounded Intelligence works both sides: Sven Hoffmann is a senior environmental practitioner of three decades who uses AI daily in live project delivery and co-chairs the Australasian Land and Groundwater Association’s Special Interest Group on AI. We advise, design and validate; enterprise builds sit with trusted partners, and we keep the selection vendor-neutral.

What does an AI readiness review cover?

Four things: what AI is already in use across the organisation, including the tools nobody registered; where the useful opportunities sit, ranked by value and risk; what has to be true before those proceed, across governance, data and skills; and a 90-day plan scoped to what the organisation can actually resource. The output is a costed, sequenced plan rather than a maturity score.

How do engagements start, and what do they cost?

Most start with a thirty-minute conversation about where AI meets your obligations and the most useful place to begin. From there the usual shapes are a briefing or seminar, a readiness sprint of four to six weeks, a governance package covering policy, validation standard and tool inventory, or a monthly advisory retainer. Each is quoted as a fixed fee against a defined scope, so the cost is known before the work starts.