Welcome to the fifteenth edition of our quarterly Digital Governance, Cyber and Privacy newsletter.

This quarter highlights a significant shift towards stronger enforcement and growing legal complexity across cyber, privacy and AI. Australian regulators continue to drive accountability through major penalties, updated privacy guidance and reforms to critical infrastructure laws. Recent developments also underscore increasing scrutiny of data practices and the evolving risks associated with AI use, including legal privilege, confidentiality and governance.

Globally, courts and regulators are testing the boundaries of AI use in legal proceedings, data governance, financial liability and consumer protection. At the same time, developments across cryptocurrency, cyber risk and emerging threats illustrate the expanding legal and operational challenges facing organisations.

Below is a roundup of key developments from Australia and around the world.

Medibank Private Limited v McClure [2026] FCAFC 38 

This March 2026 decision reinforces and arguably sharpens, the increasingly narrow pathway for maintaining legal professional privilege over cyber incident investigation reports. In a similar vein to the Federal Court decision on Optus' cyber breach, the Court emphasised that the mere involvement of external solicitors and the anticipation of litigation will not establish that the dominant purpose of a post-incident report is the provision of legal advice. Rather, the inquiry remains an objective assessment of the whole factual matrix and particularly the multiple and competing purposes typically engaged in a cyber response, ranging from regulatory compliance and governance oversight to public assurance and operational remediation.

Critically, the Court accepted that, in the context of a large-scale cyber breach, investigation reports may simultaneously serve legal, regulatory, operational and reputational functions, but found that where these non-legal purposes are substantial, they may displace legal advice as the "ruling, prevailing or most influential" purpose. In particular, the integration of the Deloitte reports into Medibank’s broader governance framework, engagement with regulators such as APRA and public positioning of the review as part of a transparency and "lessons learned" exercise were all treated as powerful indicators that the reports were not created predominantly for legal advice. This mirrors the reasoning in the Optus decision, where public communications and the framing of the review as a learning and remediation exercise undermined claims of privilege.

Do AI notetakers breach legal privilege? 

In United States v Heppner, the Southern District of New York held that a criminal defendant’s exchanges with Anthropic’s Claude were not protected by attorney-client privilege or the work product doctrine, even though the defendant said he used the tool to help prepare for discussions with his lawyers. The court found the communications were not confidential, were not between lawyer and client and were not created at counsel’s direction. The article argues this approach is too categorical and fails to account for AI functioning as a tool within the lawyer-client relationship.

The same day as the Heppner decision, the United States District Court for the Eastern District of Michigan in Warner v Gilbarco, Inc. reached a different conclusion, refusing to compel production of a litigant’s use of generative AI tools (including ChatGPT). It held that such material was either not relevant/proportionate or protected by the work product doctrine because it reflected internal legal analysis and mental impressions. The Court also rejected arguments that using AI waived privilege, emphasising that generative AI tools are "tools, not persons" and do not constitute disclosure to an adversary.

The question of AI notetakers and legal privilege is a growing topic of interest. The Australian Financial Review discussed the growing legal risks of using AI notetakers in meetings, particularly because they can create detailed transcripts that may later be discoverable in litigation or investigations. It points to the unresolved but significant risk that involving an AI tool in legally sensitive discussions could undermine legal professional privilege, especially where third-party providers may access or retain transcript data. In short, organisations should treat AI meeting tools as a governance, privacy and privilege issue, not just a productivity tool and carefully assess when their use is appropriate, particularly for board, executive and legal discussions.

Is there a breach of client confidentiality and waiver of legal privilege by using public AI?

In UK v Secretary of State for the Home Department ([2026] UKUT 81 (IAC)), the Upper Tribunal considered the increasing use of generative AI in legal practice and its role in producing fictitious case citations in tribunal proceedings. The Tribunal held that legal practitioners remain ultimately responsible for the accuracy of all material filed, regardless of whether work is delegated or generated with AI assistance and emphasised that failures in supervision may attract regulatory referral. The decision also introduced procedural changes requiring practitioners to certify that cited authorities exist and support the propositions relied upon. 

Consultation on the third phase of changes to the Security of Critical Infrastructure Act 2018 (SOCI Act)

The Department of Home Affairs has consulted on reforms aimed at improving the agility and clarity of the regime, with proposed amendments to ministerial directions powers and enhancements to the Critical Infrastructure Risk Management Program Rules, with amendments to apply to a range of asset classes, including the energy market operator, electricity, gas, liquid fuel, water, broadcasting, domain name systems, freight services and freight infrastructure.

Is your car spying on you?

The Australian Financial Review reports that the Australian Security Intelligence Organisation (ASIO) has warned politicians and public servants not to discuss sensitive information in vehicles, particularly internet-connected vehicles, due to the risk of surveillance, information theft and data collection. The BBC has highlighted that modern connected vehicles act as “computers on wheels”, collecting extensive personal data including location, behaviour and biometric indicators, often without users fully understanding the scope of data capture. According to a 2023 study by Mozilla, an analysis of the privacy policies of 25 car brands found that each one failed to meet Mozilla's privacy and security standards.

Rental unfair and excessive collection of personal information

On 22 April 2026, the Privacy Commissioner issued a determination which found that the Rent2Apply rental technology platform operated by InspectRealEstate (IRE), collected excessive personal information in breach of APP 3. The Commissioner found that IRE contravened APP 3.2 by collecting personal information that is not reasonably necessary for its functions or activities and APP 3.5 by collecting personal information by unfair means. The Commissioner emphasised the need for other RentTech providers to adapt their practices.

Updated APP 3 guidelines

APP 3 regulates when and how organisations and agencies may collect “solicited” personal information, requiring that collection be reasonably necessary for their functions or activities and carried out by lawful and fair means. The Office of the Australian Information Commissioner has updated its guidance on APP 3 to reflect modern data practices and recent regulatory decisions, including new examples covering AI, facial recognition, data scraping and tracking tools and clarifying concepts such as proportionality, 'fair means', 'reasonably necessary' and third‑party liability.

Privacy Awareness Week 2026

Explore key privacy principles in our Privacy Awareness Week video series designed to help businesses strengthen their privacy practices through practical, expert-led tips.

Vulnerabilities to gain access to one-time verification costs used in MFA

The Australian Communications and Media Authority has continued its crackdown on mobile number fraud. Recent enforcement action against SpinTel Pty Ltd identified system vulnerabilities that allowed scammers to intercept one‑time verification codes used in multi‑factor authentication, resulting in unauthorised number transfers and financial loss to consumers, while Yomojo Pty Ltd was formally warned for failing to adequately inform customers about fraud reporting obligations.

Cyber security professionals become the threat they had worked against

Two individuals who worked as ransom negotiators have pleaded guilty and been convicted of using their tech skills to participate in a spree of ransomware attacks.

World first SMS blasters

Canadian police uncovered the use of a device that mimics a legitimate cell tower to trick thousands of devices into connecting to it, allowing scammers to send messages that appear to come from trusted organisations such as banks or service providers.

Guilty to creating deepfake sexual images

A South Australian teenager has pleaded guilty in what is reported as Australia’s first conviction under new laws targeting non-consensual AI-generated sexual images.

Why ask lawyers when you have ChatGPT?

The CEO of a video game publisher has tried to use ChatGPT to wriggle out of paying a US$250 million performance bonus. The CEO ignored the company's legal department, instead following the strategy offered by ChatGPT to take control of the studio and delay a video game’s release. A Delaware court rejected the manoeuvre, ordered the reinstatement of the studio CEO and treated the scheme as a breach of contract.

Brazilian lawyers disciplined over AI prompt injecting in court forms

Lawyers were disciplined after they had filed court documents with a hidden prompt injection designed to nudge an AI system into producing a perfunctory defence. 

Tracker of AI cases in Canada

The article links through to 388 decisions that track the risk of AI references in Canadian courts.

Canadian Bill C-22

The proposed Canadian Bill has been hotly debated as it seeks to expand lawful-access powers in ways critics say could weaken encryption, increase metadata retention and create systemic cyber vulnerabilities. Supporters argue it modernises investigative powers for police and intelligence agencies, but opponents say it risks creating backdoors and undermining privacy and security.

Waiting for cryptocurrency judgment from High Court

In March 2026, the High Court heard submissions in ASIC v Web3 Ventures, the latest test of how Australia's financial services laws apply to crypto products. The case concerns whether 'Earner', a product allowing customers to loan specified cryptocurrencies to Block Earner for a fixed interest rate, constituted a financial product requiring an AFSL under the Corporations Act. At first instance, the Court found that it did, however the Full Court disagreed. ASIC has taken the matter to the High Court. Judgment is reserved.

Penalty of AU$14m ordered in ASIC v BPS Financial Pty Ltd

BPS Financial Pty Ltd has been ordered to pay $14 million after the Federal Court found it carried on a financial services business without an AFSL between January 2020 and mid-2023, issuing over 96,000 Qoin Wallets, providing financial product advice and publishing false and misleading representations about the Qoin Token's exchangeability, deriving over AU$42 million in revenue from token sales.   

UK clarifies remedies available for misappropriated crypto assets

The UK High Court has declined an opportunity to extend tortious remedies available to victims of cryptocurrency theft, finding that the property law torts of conversion and trespass to goods (which require interference with physical possession) could not be stretched to cover digital assets. The case, Ping Fai Yuen v Fun Yung Li, involved the alleged transfer of over 2,300 Bitcoin worth up to £180 million from a cold wallet by the claimant's estranged wife and sister. The decision confirms that victims of crypto fraud in the UK are left with a narrower remedies' toolkit than victims of ordinary theft and must instead rely on equitable remedies when seeking to recover stolen crypto assets.

New laws for digital asset platforms and tokenised custody platforms

The Corporations Amendment (Digital Assets Framework) Act 2026 received royal assent on 8 April 2026, bringing digital asset platforms (DAPs) and tokenised custody platforms (TCPs) within the purview of Australia's existing financial services licensing regime for the first time. Operators of these platforms will be required to hold an Australian Financial Services Licence (AFSL) and comply with the obligations that apply to all licence holders under the Corporations Act, in addition to the new digital asset‑specific requirements introduced by the Act.

41 crypto kidnappings in 2026 in France

Violent attacks targeting cryptocurrency holders have surged in France, with authorities recording more than 41 kidnappings or hostage-takings in the first few months of 2026. France now accounts for roughly 70% of known "wrench attacks" globally (cases where physical violence or kidnapping is used to force victims to transfer digital assets), with targets including crypto executives and their families.

Can you get a patent for a system that uses an Artificial Neutral Network?

In the February 2026 Appeal, Emotional Perception AI Limited (Appellant) v Comptroller General of Patents, Designs and Trade Marks (Respondent) [2026] UKSC 3, the UK Supreme Court considered whether a UK patent can be obtained for a media file recommendation system which uses an artificial neural network (ANN). The Court found that:

• a “problem and solution” approach to Art 52 of the European Patent Convention (EPC) (which excludes from patentability “programs for computers … as such”) is not the only way of assessing invention nor the only way of factoring the exclusions in Art 52(2) into the overall consideration of patentability;
• an ANN is a “program for a computer” within the meaning of EPC Art 52(2)(c); and
• in considering whether the entire subject matter of the claims were excluded, it is open to the UK Intellectual Property Office and UK courts to adopt appropriate methods of identifying the technical character of the invention, as the necessary first part of whether an invention is novel (Art 54 EPC) and an inventive step (Art 56 EPC).

Using copyrighted song lyrics violates German copyright law

The Regional Court of Munich has issued a landmark judgment in GEMA v OpenAI (Case No. 42 O 14139/24), finding that the use of copyrighted song lyrics for training generative AI models without a licence violates German copyright law. The Court found that responsibility lies with OpenAI as the operator, who makes decisions on architecture and dataset selection, not end-users and that text and data mining exceptions cannot justify outputs that replicate original works.

Victoria's judicial guidelines on AI use

The Supreme Court of Victoria has issued a new practice note and judicial guidelines permitting AI use but holding practitioners responsible for the accuracy and relevance of information in court documents. The practice note cautions court users to be aware of the risks of using AI to complete court documents, privacy considerations when using AI, possible costs orders for filing inaccurate court documents and potential sanctions for reliance on unverified AI outputs. Judicial officers are prohibited from using generative AI for decision-making. 

Quebec Superior Court set aside hallucinated award

In April 2026, a Canadian Superior Court struck down a Montreal arbitrator's August 2025 decision, which sought to resolve a dispute between a health authority and a medical clinic after finding the arbitrator relied on non‑existent legal cases generated by AI hallucinations. The Court referenced the Canadian Judicial Council rules, which prohibit judges from delegating decision-making powers to AI and cautioned that legal professionals are obligated to ensure the accuracy of proceedings.

Mass credential harvesting imbedded in Claude

A threat actor used Anthropic’s Claude Code AI assistant in a large-scale credential harvesting operation that exploited a critical react vulnerability and compromised over 900 organisations with 13,000 files and credentials compromised. Researchers found the AI tools were used mainly to assist the threat actor's development and troubleshooting workflow rather than directly executing attacks.

Capping cyber payouts over AI and LLM jacking

The Financial Times reported that insurers are introducing new caps on payouts for cyber losses and regulatory fines related to AI use, as the industry rushes to reduce its exposure to the rapidly advancing technology.

AI images being used to scam insurance

Insurers have noticed an increase in the number of reported claims that appear to be supported by AI manipulated images and fabricated documents.  

Reporting windows for cyber claims in Ohio, USA

The United States District Court for the Southern District of Ohio, applying Ohio law, held that a negligence claim arising from a payment redirection fraud was not first made or reported while the first cyber policy was in effect and that notice was outside the second cyber policy’s 30-day reporting window. The court also held that a prior knowledge exclusion barred coverage for any cyber incident or Wrongful Act, prior to the policy’s effective date.

*Note: for some publications, you may require a current subscription to read the full article.