# Regulatory AI > Hiaitus builds AI systems that encode a regulatory framework and screen work against it before publication or sign-off. Any industry, any written rules. Source: https://www.hiaitus.ai/services/regulatory-ai Compliance systems that read your codes, your policies, and how regulators actually ruled. _Regulatory AI_ ## In your business is a document that says what you can't do, lost on a network drive, until something goes wrong. At Hiaitus we build simple, high impact AI systems that learn the codes you answer to and the policies you write yourself, then automatically check work against both before anything is actioned or signed off. We built one for Australian advertising compliance and it is in market today. The method works on any codified framework. | Focus Area | Pre-Hiaitus | Post-Hiaitus | | --- | --- | --- | | The codes | In someone's head | Encoded once | | Your policies | Usually skipped | Checked every time | | Past decisions | If anyone remembers | Matched automatically | | Sign off | Your call | Still your call | **Human** In control, always. Hiaitus AI platforms flag the issues, approvals stay human. - [Get a feasibility assessment](https://www.hiaitus.ai/contact) - [See Checked It](https://www.checkedit.ai) ## What is a regulatory AI system? A regulatory AI system reads a written framework, turns its requirements into tests, and checks work against those tests before anything is published or signed off. Hiaitus builds them in Australia. Checked It, our own product, does this for advertising across six sectors against codes from the ABAC Scheme, Ad Standards, the Therapeutic Goods Administration, ACMA and FSANZ. The same method works on Fair Work awards, the Australian Privacy Principles, NDIS Practice Standards, or a policy nobody has opened since 2023. _How we build_ ## How does a regulatory AI system actually work? Three moves. The same three we use on every build, pointed at a framework instead of a workflow. ### 01 · The rules, and how they were applied _Ingest_ Public codes, internal policy, and the decisions in between. We take the public code, and the parts nobody publishes: your brand guidelines, your legal team's standing positions, the two-page list of things you would never do. Then the decisions. Every case a regulator has ruled on is a worked example of how the rule reads in practice. ### 02 · Rules become tests, precedent becomes reference _Encode_ Where a rules engine stops and this starts. A checklist tells you a claim needs substantiation, which your reviewer already knew. It cannot tell you that a near-identical claim was dismissed in 2022 and a bolder one upheld in 2024. We encode the rule as a test and the decision as a reference case, so the system reasons about the gap between them. ### 03 · Assessed before it goes out, decided by a person _Screen_ A faster first pass. Your reviewer still owns the call. Work gets screened before it reaches legal, the client or the regulator. Findings come back with the clause cited and the closest decided case attached, so your reviewer confirms a position rather than starting from a blank page. The system never signs anything off. That stays with a named person, and the log records who. ## Which frameworks can we build AI systems for? Any framework with written rules. Where the regulator publishes its decisions or guidelines, we encode those as well, so the system knows how a rule needs to be applied and how it has been applied in practice, inclusive of what you can (as opposed to cannot) do. **Framework** | **Who decides** | **We check against** **Advertising and marketing** _Built: Checked It_ | Ad Standards, ABAC Scheme, ACMA, TGA | Rules + precedent Financial product promotion | ASIC | Rules + precedent Workplace relations and Modern Awards | Fair Work Commission | Rules + precedent Privacy and the Australian Privacy Principles | OAIC | Rules + precedent NDIS provider compliance | NDIS Quality and Safeguards Commission | Rules + precedent Aged care quality standards | Aged Care Quality and Safety Commission | Rules + precedent Work health and safety | Safe Work Australia | Rules + precedent Telecommunications consumer protections | ACMA | Rules + precedent Franchising disclosure | ACCC | Rules + precedent Energy retail conduct | Australian Energy Regulator | Rules only Food labelling and composition | FSANZ | Rules only Mandatory climate reporting | ASIC, AASB | Rules only **Your own internal policy** _Most of what we build_ | You | Rules only --- Not on the list? The only question is whether your rules are written down. If they are, we can encode them. > Some frameworks we will not take on. Clinical decision support, personal financial advice, tax agent services and migration advice each carry a licence obligation that would land on us rather than on you. Anti-money laundering is a firmer no: thin enforcement record, and the tipping-off provisions make it genuinely risky. - [Get a feasibility assessment](https://www.hiaitus.ai/contact) ## What does one of these look like in market? Checked It is ours. It screens advertising creative against Australian regulatory codes before publication, across six sectors, and it is in market today. It is the reason this page exists. We did not write a capability deck about regulatory AI and then go looking for a client. We built the system, sold it, supported it, and found out where it breaks. The engines are versioned, so when a regulator moves the prompts move. ASIC reissued its advertising guidance for financial products in June 2026, the first substantive update in over a decade. A tool that was correct in March can be quietly wrong by September, and keeping one current is the part most vendors do not price. Checked It has its own pricing and its own trial at checkedit.ai. Here it is doing one job, which is showing you we have already done this once. ### Codes built into the engine - Alcohol: ABAC Scheme, Ad Standards, AANA - Health and Supplements: TGA, Ad Standards, FSANZ - Gambling and Gaming: ACMA, Ad Standards, AANA - Food and Beverage: FSANZ, Ad Standards, AANA - News and Publishing: Australian Press Council, ACMA, MEAA - Insurance Brokers: ASIC, ACCC, Insurance Council of Australia _Q3 2026_ ## What are we building next, and what can we build into yours? Three capabilities are in build inside Checked It this quarter. Each one is a pattern that transfers to any framework. ### Q3 · Brand guideline screening _In build_ The rules that are yours, not the regulator's. Every organisation has a document nobody opens. For a brand it is tone, colour, the words you avoid, and the short list of things you would never do. It gets briefed once, and then an influencer posts on a Saturday. We are building screening that reads those guidelines the way we read a regulatory code, and the same pattern handles an internal policy or a licence condition. Regulatory breaches are rare; off-brand work happens every week, and it is what your client notices first. ### Q3 · Pre-approved claims libraries _In build_ Stop re-approving claims you cleared last year. Most regulated organisations keep a spreadsheet of claims legal has already signed off, by product, by market, sometimes by date. It is rarely consulted, because checking it is slower than asking. We are building ingestion for those files so cleared claims are recognised on sight and lifted out of the queue, leaving your reviewer only what is genuinely new. ### Q3 · Precedent comparison _In build_ How was something like this judged before? Put up a piece of work and see the decided cases closest to it in substance, with the outcome and the reasoning. Useful when a reviewer is deciding, and more useful in the conversation afterwards, when someone senior asks why the answer was no. _FAQ_ ## Questions we get asked ### What is a regulatory AI system? A regulatory AI system reads a specific written framework, turns its requirements into tests, and screens documents, creative or decisions against those tests before they are published or signed off. It differs from a general AI assistant because the rules are encoded rather than recalled, so the same input produces the same finding every time. Where past decisions on that framework are public, the system also holds them as reference cases and compares new work against what has already been judged. The output is a flagged assessment with the clause cited, and a person still makes the decision. ### Which Australian advertising codes can AI pre-screen against? Our product Checked It screens across six regulated sectors. Alcohol, against the ABAC Scheme and Ad Standards. Health and supplements, against the Therapeutic Goods Administration, Ad Standards and FSANZ. Gambling and gaming, against ACMA and Ad Standards. Food and beverage, against FSANZ and Ad Standards. News and publishing, against the Australian Press Council, ACMA and the MEAA. Insurance brokers, against ASIC, the ACCC and the Insurance Council of Australia. The AANA Code of Ethics applies in parallel across every sector. ### Can AI approve regulated content on its own? No, and we would not build a system that claimed to. Approval is a decision with a person's name attached, and a regulator investigating a breach will want to know who made it and on what basis. What AI does is the first pass: read the work against the framework, flag what needs attention, cite the clause, and attach the comparable decided case where one exists. Your reviewer then confirms a position instead of starting from a blank page. ### Who is liable if AI approves non-compliant work? The organisation publishing the work, and increasingly the agency alongside them. Australian regulatory decisions name the parties responsible, and using a screening tool does not transfer that responsibility to the tool's vendor. This is one reason we build systems that flag rather than approve, and that record who signed off. Treat any tool offering to take the decision off your hands as a commercial risk. We are not lawyers and this is not legal advice, so if liability allocation is a live question for you it belongs with your legal team before a tool is chosen. ### What evidence does a regulator expect if AI was in the approval chain? Regulators are generally concerned with the outcome and the process behind it rather than the software involved. In practice that means showing what was checked, against which version of which framework, on what date, and who accepted the result. We build that logging as a core part of the system rather than an add-on: the assessment, the clauses applied, the ruleset version in force at the time, and the human sign-off. A decision you cannot reconstruct six months later leaves you worse off than no tool at all. ### How is this different from a rules engine or a compliance checklist? A checklist encodes what the rule says. It will tell you a therapeutic claim needs substantiation, which your reviewer already knew. What it cannot tell you is where the line has actually been drawn, because that lives in the decisions rather than the code. A near-identical claim dismissed in one year and a bolder one upheld in another is the information a reviewer needs, and it only exists in the adjudications. We encode both: the rule as a test, the decision as a reference case, and the system reasons about the distance between them. ### Does my framework have enough public precedent for this to work? It might not, and we will tell you before you spend anything. Screening works on any framework that is written down, including your own internal policy. The precedent layer is the part that needs public decisions. Advertising has an unusually good record, freely available. Fair Work, the OAIC, the NDIS Commission and the Aged Care Quality and Safety Commission all publish decisions in usable volume. Food labelling has a strong standard and no central database of decisions. Mandatory climate reporting has no decision record at all yet. The feasibility assessment establishes which of those situations you are in. ### Should we build our own or buy an existing platform? Buy, if a platform already covers your framework. Checked It exists because Australian advertising codes are common to enough businesses to make a product worth building. Build, when the rules are yours: an internal policy, a set of brand guidelines, a library of pre-approved claims, a licence condition specific to your operation. No vendor will ever cover those, because there is no second customer for them. The practical answer is usually both, a platform for the public codes and something bespoke for the rules only you have. ### How long does a regulatory AI build take, and what does it cost? Most first builds run in weeks rather than quarters. We scope tightly so you get a working version early and extend it once it proves out. Cost depends on the framework: how large it is, whether the decisions are machine-readable, and how much of your internal policy needs structuring before it can be encoded. We give a fixed quote after the feasibility assessment rather than a range up front, because the range would be so wide as to be useless. You will know the number, and what it buys, before any work starts. ### Does using an AI screening tool guarantee we will not breach a code? No. Nothing does, including a careful human review, and any vendor telling you otherwise is selling you a risk you cannot see. Codes are interpreted by panels applying judgement to context, and reasonable people disagree about where a line sits. A screening system catches the clear problems early, surfaces the arguable ones with the relevant precedent attached, and gives your reviewer a documented basis for the call. Expect fewer avoidable breaches, not immunity. Something else? Ask us in the feasibility assessment. It's free, and it quite often ends with us saying the method won't work for you. _Start here_ ## Tell us what you have to comply with. We'll tell you if this works. A feasibility assessment is a short conversation and a written answer: whether your framework can be encoded, whether the precedent layer is available to you, what a first build would cover, and what it would cost. If the answer is that it will not work, you get that in writing too, with the reason. - [Get a feasibility assessment](https://www.hiaitus.ai/contact) - [hello@hiaitus.ai](mailto:hello@hiaitus.ai)