Deepfakes and cloned voices: how property identity checks respond

In this article

Written by

Kooky

Builder of Shaka, the payment router that pays every agent their commission on closing date.

About Kooky and Shaka →

Property transactions run on recognition. A solicitor recognises a client's face against a passport, an agent recognises a vendor's voice on the phone, a lender recognises a payslip as the kind of document an employer issues. Each of those acts of recognition was designed in a period when a face, a voice and a document were hard to imitate well. Generative artificial intelligence has changed the cost of imitation, and the published guidance of Australian courts, regulators and the body that writes the electronic conveyancing rules has begun to say so.

This guide looks at one narrow question: what synthetic media can imitate, and how the checks built into a Queensland property dealing respond to it. It covers what Queensland's courts and the corporate regulator have published, how verification of identity is designed under the national conveyancing rules and why its central element is a meeting in person, and the checks that published sources describe for buyers, sellers and agents.

What "synthetic media" covers

Queensland's courts give a working definition. Their guidelines for judicial officers on the use of generative AI, in the revised version issued on 15 September 2025, describe generative AI as a form of AI that enables users to generate new content, which can include text, images, sounds and computer code.

Three kinds of output matter for property. A cloned voice is audio generated to sound like a particular person. A deepfake is an image or video generated or altered to show a particular person, sometimes live on a call. A synthetic document is a text or image file made to look like something an institution issued: an identity card, a statement, a letter on letterhead.

The same guidelines state the consequence in a single passage. AI tools, they say, are now being used to produce fake material, including text, images and video, and judges should be aware of the challenges posed by deepfake technology. The passage adds a piece of perspective that is useful outside a courtroom too: courts have always had to handle forgeries, and allegations of forgery, of varying levels of sophistication.

Voices: what has been shown and what has been reported

The clearest Australian account of voice cloning written for consumers is an investigation published by the consumer group CHOICE on 25 August 2024. It reported that a convincing clone can be built from mere seconds of original audio, that videos posted on social media are one source of such audio, and that its own journalists were able to clone voices with a web-based tool for a subscription of US$1.66 a month.

The same article supplies a necessary counterweight. The National Anti-Scam Centre, which sits within the ACCC, told CHOICE it had received fewer than five reports of suspected AI voice cloning since 2022. National Australia Bank, the article noted, had nonetheless listed AI voice clones among the scams it expected to emerge in 2024. CHOICE also described a call received by a Sydney advertising executive in which a voice imitating the Queensland Premier of the day promoted an investment scheme; the recipient is reported as finding the imitation good, if slightly robotic.

Those figures are two years old at the date of this guide and describe reports, not incidents. Many people who receive a strange call do not report it, and a person deceived by a good imitation may never learn that the voice was generated. What the CHOICE reporting establishes is capability and price: the ability exists and it is cheap. It does not establish how often the ability is used against property transactions, and no source read for this guide gives such a number.

Faces, video and documents on a screen

For images and video, the most relevant published statement comes from the Australian Registrars' National Electronic Conveyancing Council, known as ARNECC, which writes the Model Participation Rules for electronic conveyancing. Its Guidance Note 2 on verification of identity, in the version updated in August 2024, deals with whether a video call can stand in for a meeting.

Its answer has two parts. A video call does not satisfy the Verification of Identity Standard, because the Standard requires an interview with both people physically present and the sighting of original documents. And where a practitioner chooses to verify identity in some other way and uses video as part of it, the guidance note recommends caution, giving as its reason that video technology may be manipulated or forged.

That sentence was published two years before this guide. It is a plain acknowledgment, by the rule-maker for conveyancing, that an image on a screen is evidence of a lower grade than a person in a room.

The third category is the synthetic document. Queensland's judicial guidelines mention fake text and images alongside video, and they give a second, related warning about expert evidence: judges are asked to be alert to the use of AI by experts in generating or expressing an opinion, and to consider requiring the expert to identify precisely how AI was used.

In a property dealing the documents at risk are the ordinary ones. Identity documents presented at a verification. Letters of authority. Evidence that someone is entitled to deal with the land, such as a rates notice. Financial documents supplied to a lender. Each has traditionally been assessed partly by appearance: the layout, the logo, the typeface, the feel of the paper.

ARNECC's guidance note points the assessment elsewhere. Among the further steps it lists where doubt arises are using electronic verification services for Australian documents, and, for a foreign document, checking that it looks the same as the examples on the issuing country's government website. Both are checks against a source outside the document. A file can be made to look perfect. It cannot make an issuing authority's records agree with it.

What Australian authorities have published

Published material on synthetic media is spread across bodies with different jobs, and none of it is written for property specifically. The table sets out what was read for this guide.

The table is not a complete list of what Australian agencies have said.

ASIC's open letter of 8 May 2026

The Australian Securities and Investments Commission regulates the licensees that sit behind a property purchase: lenders, mortgage brokers and other financial services businesses. On 8 May 2026 it published an open letter addressed to all licensees and market participants. According to a summary published that month by the regulatory news service Regulation Tomorrow, the letter said that frontier artificial intelligence intensifies the global cyber risk environment and that AI is accelerating the discovery and exploitation of vulnerabilities.

The letter, on that account, described cyber resilience as a core licensing obligation and set out twelve actions. They include reassessing cyber plans against current risks, confirming governance and decision-making arrangements, identifying and protecting critical assets, reviewing user access and privileges, patching promptly, layering defences, preparing incident response and business continuity plans, managing third-party risk, and using AI defensively. The summary attributes to ASIC Commissioner Simone Constant the message that cyber risk management starts with boards and executives.

Two things should be said about its relevance here. The letter, as summarised, is about systems. It does not mention deepfakes, impersonation or identity verification. Its connection to synthetic media is indirect: a lender or broker whose systems are compromised is a source of the genuine documents and personal details from which a convincing imitation is assembled. The second point is that it applies to ASIC's licensees. Real estate agencies and law practices are not addressed by it unless they also hold such a licence.

The corporate plan of 26 August 2026

ASIC named deepfakes directly three and a half months later. Its media release of 26 August 2026, announcing the Corporate Plan 2026-27, says the regulator will examine the use of AI in banking customer services, the potential effects on consumers and investors, and AI-driven manipulation, deepfakes and misinformation as they affect market integrity.

The release lists scams, with debt collection, among the consumer protection areas where harm spreads quickly, and commits ASIC to building its own data, intelligence, AI and cyber capabilities. ASIC's chair, Sarah Court, is quoted in the release to the effect that AI can improve services and decision making but must not weaken accountability or the protections owed to consumers and investors. An earlier ASIC release in 2026, numbered 26-195MR, carries a title warning that scammers are using AI to spin vast webs of deception.

For a property reader the plan is context, not instruction. It records that the national financial regulator now treats deepfakes as a supervisory subject. It sets no rule for a conveyance.

How verification of identity is designed

The rule that does govern a conveyance is in the Model Participation Rules, which bind the solicitors and lenders who subscribe to an electronic lodgment network. ARNECC's Guidance Note 2 explains the design.

The obligation is to take reasonable steps to verify identity, measured by what an ordinarily prudent subscriber or mortgagee would have done in the circumstances. It covers clients, mortgagors, persons to whom certificates of title are provided, and the people who sign and administer on the subscriber's behalf.

The rules then offer a defined route. Schedule 8 contains the Verification of Identity Standard. If the Standard is properly carried out by the subscriber, by a mortgagee, or by an identity agent appointed in writing, the subscriber is deemed to have taken reasonable steps. The Standard is not compulsory. A position statement published by ARNECC in April 2026, dealing with clients overseas who cannot reach a consulate, confirms that a subscriber may verify identity in some other way that amounts to reasonable steps, and must then be able to show why the steps were reasonable.

The Standard itself has four features that bear on synthetic media.

• An interview in person. The verifier must conduct a face-to-face in-person interview, with both people physically present. The guidance note says in terms that video technology such as Skype or FaceTime does not qualify.
• Original, current documents. The verifier must carefully inspect the documents and ensure they are original and current. The one exception is an Australian passport that expired within the last two years.
• A likeness test. Any photograph on the documents must reasonably correspond with the appearance of the person being identified.
• The highest category available. Documents are ranked in categories, and the highest category the person can produce must be used. The lowest category for Australian residents relies on an identifier declaration, a statutory declaration by someone who knows the person, whose own identity must be verified under the Standard.

Evidence supporting the verification is kept for seven years from lodgment. A verification done within the previous two years need not be repeated, provided the subscriber takes reasonable steps to ensure it is dealing with the person identified earlier. And the subscriber certifies, in the registry instrument, that reasonable steps were taken.

Why the in-person element matters more now

Set those four features against the three kinds of synthetic output and the logic of the design becomes visible.

A cloned voice has no role in an in-person interview. A live deepfake depends on a camera and a screen between the two people, and the Standard removes both. A synthetic document delivered as a file is not an original that can be handled, and the Standard asks for originals. What remains possible under the Standard is the older risk it was written for: a physical forged document, or a person who resembles the photograph on a genuine one. The likeness test and the document categories are aimed at those.

Where verification is done by another route, the comparison changes. The April 2026 position statement lists, among the alternatives, electronic verification services for Australian documents and conducting the verification by electronic means with further steps. A separate ARNECC guidance query on verification of identity adds that customer due diligence or know-your-customer processes may amount to reasonable steps. These routes exist because not every client can attend an office, and the rules permit them. They also place a screen back between verifier and client, which is why the guidance note attaches its warning about manipulated video to exactly this case and asks the subscriber to be able to justify the method in the circumstances of the particular verification.

The table summarises the difference.

When doubt arises: the further steps

The guidance note does not treat a completed checklist as the end of the matter. Further enquiries should be made, it says, where doubt arises or should reasonably have arisen about a transaction and a person's identity.

It names the triggers: an identity document that does not appear genuine, a photograph that is not a reasonable likeness, a person who does not appear to be the person the documents relate to, or any other circumstance that makes further steps reasonable. Its examples are practical. A client with very few identity documents and no explanation. Documents the verifier learns have been cancelled. A signature on the identity document that differs from the one on the client authorisation or mortgage.

It also lists circumstances around the transaction, apart from the documents, that should raise attention: urgency, doubt about the instructions, and a non-standard mortgage over an unencumbered title. Read in sequence, the guidance amounts to a three-stage response.

None of these steps depends on detecting that an item is synthetic. That is their strength. A verifier is not asked to out-analyse the software. The verifier is asked to notice that something does not fit and to go and check it somewhere else.

Checks described for buyers, sellers and agents

The published advice for people outside the legal profession follows the same principle, expressed for everyday use. CHOICE's 2024 article sets out the checks most often recommended against a cloned voice.

The first is a prearranged codeword, or an agreed question and answer, between people who expect to exchange important instructions. The second is to end an unexpected call and ring the person back on a number already held on file or in a phone's address book, never on a number supplied during the call. The third is to treat with suspicion any call from an unknown number that claims to be from a familiar person. The fourth concerns supply: adjusting social media privacy settings limits how much recorded speech is publicly available.

The article also lists the signs that a call may not be what it seems: a sense of urgency, an unwillingness to explain further, the absence of normal social cues or of the greetings and turns of phrase the real person uses, and an inability to keep up a spontaneous conversation.

Carried into a property transaction, those checks have obvious places to sit. A seller and an agent who will exchange instructions by phone during a campaign can agree a word at the listing appointment. A buyer who receives a call that changes an arrangement, whatever it is, can ring the firm's published switchboard number before acting. An agent taking instructions from an owner never met in person has reason to seek confirmation through a second channel that was established before the instruction arrived. Each of these is a small procedure, and each works without anyone needing to judge whether a voice sounds real.

For documents, the equivalent habit is to confirm with the issuer. A letter that appears to come from a law firm, a bank or a body corporate manager can be confirmed with that organisation through details obtained independently of the letter.

What the published material leaves open

Three gaps are worth stating plainly, because a guide that hides them would overstate what is known.

There is no published count of synthetic media being used in Queensland property dealings among the sources read here. The only Australian figure found, the National Anti-Scam Centre's fewer than five reports of voice cloning, dates from August 2024 and covers all scam types.

The conveyancing rules were not rewritten for deepfakes. Guidance Note 2 predates the current generation of tools and addresses manipulated video in one sentence. Its protection comes from a design choice made for other reasons, the in-person interview, and that protection applies in full only when the Standard is the route used.

And the regulator statements of 2026 are addressed to financial licensees and to markets. ASIC's letter and corporate plan show the direction of official attention. They do not set a standard of conduct for a real estate agency or a law practice handling a sale.

What the sources do agree on is the direction of the answer. Queensland's judges are told to expect fabricated material and to ask what checks were made. Conveyancing practitioners are told that a screen is weaker evidence than a room, and to look outside a document for confirmation of it. Consumers are told to call back and to agree a word in advance. All three rest on the same idea: when anything can be made to look or sound genuine, confidence has to come from a second, independent source.

Kooky, from Shaka

Kooky edits Queensland Estate and builds Shaka, the payment router he made for Queensland property professionals. One payment comes in, and every agent, agency and party in the deal receives their signed share on closing date.

About Kooky and Shaka →