AI regulatory and policy gaps

October update: WA Police listed further deployments for 13, 14 and 16 October. Its published trial results, updated on 6 October, recorded 905,117 faces scanned, 209 alerts and eight incorrect alerts. Published trial results.

WA Police is trialling live facial recognition in public locations. Cameras detect faces in their field of view, create biometric templates and compare them against an alert list. The alert list can include wanted people, suspects in serious offences, missing people, people subject to restrictions and people police consider may pose a risk to themselves or others.

Why it is concerning

Live facial recognition allows people in public spaces to be biometrically scanned without individually agreeing to it. Key concerns include false matches, watchlist criteria, disproportionate impacts on particular groups, and how far this type of surveillance may expand after a trial.

Read more: www.wa.gov.au/organisation/western-australia-police-force/live-facial-recognition-technology

Complain or raise a concern: WA Corruption and Crime Commission, police misconduct: www.ccc.wa.gov.au/for-the-public/police-misconduct

WA Corruption and Crime Commission, make a report: www.ccc.wa.gov.au/for-the-public/make-a-report

NSW Police uses facial recognition to generate investigative leads. Parliamentary material records hundreds of uses each year.

Why it is concerning

Facial recognition can turn ordinary government-held photographs into biometric search material. Concerns include false matches, limited public visibility over search criteria, expansion of databases and the risk that a biometric match is given more evidentiary weight than it deserves.

Read more: www.parliament.nsw.gov.au/lcdocs/other/22523/4.%20ASQ%20-%20Hon%20Yasmin%20Catley%20MP%20-%20Police%20and%20Counter-terrorism%20-%20Received%2021%20Oct%202025.pdf

Complain or raise a concern: Law Enforcement Conduct Commission: www.lecc.nsw.gov.au/complaints/make-a-complaint

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

NSW Police has acknowledged using Fivecast for intelligence purposes. Fivecast products can analyse large volumes of publicly available online information and identify connections, entities, patterns and potential risks. NSW Police has declined to publicly disclose aspects of its methodology.

Why it is concerning

Systems that analyse people's online activity can expand police surveillance beyond direct investigations. Concerns include opaque risk indicators, network or association-based profiling, monitoring of lawful activity and limited ability for a person to know how an intelligence assessment was created.

Read more: www.parliament.nsw.gov.au/lcdocs/other/22303/Transcript%20-%20PC5%20-%203%20September%202025%20-%20Budget%20Estimates%20%28Catley%29%20HIGHLIGHTED%20FOR%20QoN.pdf

Complain or raise a concern: Law Enforcement Conduct Commission: www.lecc.nsw.gov.au/complaints/make-a-complaint

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

The South Australian Auditor-General reported that SA Police uses facial recognition for intelligence purposes against material including CCTV and custody photographs, mobile phone detection technology and social-media monitoring for counter-terrorism. It also reported plans for machine-learning tools intended to identify potential threats, hidden connections and investigative leads.

Why it is concerning

This combines biometric identification, online surveillance and algorithmic intelligence analysis within policing. Concerns include association-based profiling, false matches, opaque threat indicators and limited public visibility over how intelligence systems generate or prioritise leads.

Read more: www.audit.sa.gov.au/sites/default/files/2025-09/Report%208%20of%202025%20-%20Annual%20report%20-%20Part%20A%20-%20Executive%20summary.pdf

Complain or raise a concern: Office for Public Integrity, SA Police complaints: www.publicintegrity.sa.gov.au/police-complaints

Clearview AI built a facial-recognition database by scraping billions of photographs from publicly available websites and social media. Australian police forces trialled the tool. The Australian Federal Police-led Australian Centre to Counter Child Exploitation uploaded facial images during a free trial, including in some active investigations.

The Australian Information Commissioner later found that the AFP had failed to complete the required privacy impact assessment before using Clearview AI and had not adequately assessed privacy and security risks.

Why it is concerning

The case shows how a powerful biometric surveillance product could enter Australian policing before appropriate privacy assessment had occurred. It also demonstrates the risks of government agencies sending biometric information to a private overseas company whose database was built without the consent of the people pictured.

Read more: www.oaic.gov.au/news/media-centre/afp-ordered-to-strengthen-privacy-governance

OAIC findings on Clearview AI: www.oaic.gov.au/news/media-centre/clearview-ai-breached-australians-privacy

Complain or raise a concern: Office of the Australian Information Commissioner: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

The Australian Federal Police publicly acknowledges using AI capabilities including natural-language processing, translation, optical character recognition, limited object-recognition computer vision, autonomous 3D mapping and large language model tools. The AFP says operational details may be withheld where disclosure could reveal law-enforcement capabilities.

Why it is concerning

The public can see the broad technology categories but not necessarily the individual operational use cases. That makes it difficult to independently assess how AI affects investigations, intelligence assessments or decisions about individuals.

Read more: www.afp.gov.au/sites/default/files/2025-10/AFP-AI-TransparencyStatement.pdf

Complain or raise a concern: Commonwealth Ombudsman, AFP complaints: www.ombudsman.gov.au/complaints/australian-government-agency-complaints/australian-federal-police

Privacy complaints to the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

Australia's Identity Matching Services include the Face Verification Service and the National Driver Licence Facial Recognition Solution. The infrastructure allows biometric information derived from government identity documents to be checked across participating systems. The broader Face Identification Service is not currently operational.

Why it is concerning

A national biometric matching infrastructure creates the technical capacity for government-held photographs to be used in increasingly automated identity checks. Questions remain about consent, function creep, access controls, retention and future expansion.

Read more: www.idmatch.gov.au/about-our-services

Complain or raise a concern: Office of the Australian Information Commissioner: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

NSW child protection uses Structured Decision Making tools at multiple points in a case. These tools help structure decisions about whether a report should be made, whether a matter should be investigated, whether a child can remain at home, whether a case should remain open, whether restoration should occur and whether a case can close.

Why it is concerning

Structured tools can improve consistency, but they also reduce complex family situations into predefined factors and categories. Risks include false positives, false negatives, cultural mismatch, workers becoming anchored to a risk score and historical data reproducing existing patterns of intervention.

Read more: www.dcj.nsw.gov.au/service-providers/deliver-services-to-children-and-families/child-protection-services/structured-decision-making.html

Australian Institute of Family Studies overview of child-protection risk assessment: www.aifs.gov.au/resources/resource-sheets/risk-assessment-approaches-child-protection

Complain or raise a concern: NSW Department of Communities and Justice, child protection complaints: www.dcj.nsw.gov.au/contact-us/complaints-and-feedback/child-protection-complaint-handling.html

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

Queensland Child Safety uses Structured Decision Making assessments at key points in child-protection intervention. Current policy identifies tools including safety assessment, child strengths and needs assessment and parental strengths and needs assessment. At intake, structured decision-making is used alongside professional judgement to help decide whether concerns meet the notification threshold and how quickly Child Safety should respond.

Why it is concerning

These tools can shape high-impact decisions about children and families even though they are not machine-learning systems. Concerns include rigid categorisation, false positives and negatives, cultural fit, historical bias and the possibility that a structured score becomes an anchor for later professional judgement.

Read more: www.cspm.csyw.qld.gov.au/our-approach/principles

Queensland Structured Decision Making policy: www.families.qld.gov.au/_media/documents/protecting-children/407-structured-decision-making.pdf

Complain or raise a concern: Queensland Ombudsman: www.ombudsman.qld.gov.au/how-to-complain

Office of the Information Commissioner Queensland: www.oic.qld.gov.au/community/how-to-make-a-privacy-complaint

Queensland child-safety review material has recommended examining predictive algorithms and machine learning to help identify risks, needs and protective factors, prioritise cases, allocate resources, guide referrals and synthesise case histories.

Why it is concerning

Predictive child-protection systems can attach future-risk labels to families based on administrative data. If deployed, major issues would include false positives, bias in historical child-protection data, disproportionate impact on already over-represented communities, transparency about model inputs and the ability of families to challenge an algorithmically influenced assessment.

Read more: www.parliament.qld.gov.au/Work-of-the-Assembly/Tabled-Papers/docs/5825T2048/5825t2048.pdf

Complain or raise a concern: Queensland Ombudsman: www.ombudsman.qld.gov.au/how-to-complain

Office of the Information Commissioner Queensland: www.oic.qld.gov.au/community/how-to-make-a-privacy-complaint

Victoria's privacy regulator investigated the use of ChatGPT by a child protection worker to assist with a Protection Application Report submitted to the Children's Court. The investigation found that sensitive personal information had been entered into ChatGPT and that AI-generated material included inaccurate information that materially altered the presentation of risk.

The department later reviewed hundreds of cases and identified a group of files containing indicators that ChatGPT may have been used in court reports, case notes, case plans or risk assessments.

Why it is concerning

This is a direct example of generative AI entering statutory social-work documentation. Risks include hallucinated facts, distorted risk assessments, confidentiality breaches, court reliance on inaccurate material and supervisors failing to detect AI-generated errors.

Read more: www.ovic.vic.gov.au/regulatory-action/investigation-into-the-use-of-chatgpt-by-a-child-protection-worker/

Investigation report: www.ovic.vic.gov.au/wp-content/uploads/2024/11/DFFH-ChatGPT-investigation-report-20240924-Re-upload.pdf

Complain or raise a concern: Office of the Victorian Information Commissioner: www.ovic.vic.gov.au/privacy/for-the-public/privacy-complaints/

Victoria's Rapid Child Safety Review recommended a new Shared Intelligence and Risk Assessment Capability that would bring together information from child protection, police, Working with Children Check functions and other agencies. The review specifically identified an opportunity for AI to rapidly scan information and flag patterns of concern.

The review also proposed broader use of information and intelligence in Working with Children Check decisions, including information that may not have resulted in substantiated findings.

Why it is concerning

Combining large amounts of sensitive information across agencies and using AI to identify risk patterns raises serious questions about data quality, unsubstantiated allegations, false associations, explainability, proportionality and people's ability to correct or challenge information about them.

Read more: www.vic.gov.au/rapid-child-safety-review-2025/chapter-4-overhaul-working-children-check-and-reportable-conduct-schemes-single-entity-new-risk

Government response: www.vic.gov.au/victorian-government-response-rapid-child-safety-review/reforms-led-victorian-government

Complain or raise a concern: Office of the Victorian Information Commissioner: www.ovic.vic.gov.au/privacy/for-the-public/privacy-complaints/

The NSW Human Services Dataset links de-identified administrative information about children, young people and families across government. It contains records relating to child protection, housing, justice, health, education, mental health, alcohol and other drug issues, parental risk indicators and Commonwealth welfare and medical supports.

NSW says the dataset is used to understand vulnerability, predict demand for services, guide investment and support research. It contains millions of linked records and is intended to be updated over time.

Why it is concerning

The dataset is de-identified and access is controlled, but its scale creates a powerful infrastructure for profiling and predictive research involving vulnerable populations. The same type of linked data has already been used in NSW machine-learning research predicting children's future contact with the criminal justice system.

Read more: www.dcj.nsw.gov.au/about-us/facsiar/human-services-dataset-hsds/about-the-human-services-dataset.html

Data contained in the dataset: www.dcj.nsw.gov.au/about-us/facsiar/human-services-dataset-hsds/data-contained-in-the-human-services-dataset.html

Ask questions or raise concerns: NSW DCJ Human Services Dataset contact: HSDS@dcj.nsw.gov.au

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

NSW Bureau of Crime Statistics and Research developed machine-learning models using linked government data from approximately 259,000 children. The data included justice, police, health, education, child protection, housing and revenue records. The research tested whether future first contact with the criminal justice system before age 18 could be predicted.

The researchers identified significant limitations, including false positives and poorer model performance for Aboriginal children.

Why it is concerning

The study shows that government already has the data infrastructure and technical capacity to build predictive profiles of children. If such models were ever operationalised, incorrect predictions could lead to stigma, increased surveillance or intervention based on predicted future behaviour rather than actual conduct.

Read more: www.dcj.nsw.gov.au/content/dcj/bocsar/bocsar-home/research-evaluations/2025/CJB270-Report-Predicting-first-criminal-justice-contact-before-age-18-using-a-large-linked-administrative-dataset.html

Complain or raise a concern: NSW Department of Communities and Justice complaints and feedback: www.dcj.nsw.gov.au/contact-us/complaints-and-feedback.html

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

Australian youth justice systems use structured risk and needs tools to assess young people's criminogenic needs, case-planning priorities and intervention intensity. NSW uses tools from the Youth Level of Service family.

Why it is concerning

Risk tools can encode assumptions about family structure, neighbourhood, education, peer groups and prior system contact. When historical disadvantage is treated as a predictor of future offending, the assessment process can reproduce racial, socioeconomic and institutional disparities.

Read more: www.aihw.gov.au/reports/youth-justice/youth-justice-in-australia-2024-25/contents/appendices/appendix-d-state-and-territory-systems

Complain or raise a concern: Use the complaint process of the relevant state or territory youth justice department. Privacy complaints should be made to the relevant state privacy regulator.

Corrective Services NSW developed the Domestic Violence Triage Risk Assessment Scale, or DV-TRAS, to automatically estimate the likelihood that custody-based male domestic-violence offenders will reoffend. The model uses routinely collected administrative information and was designed to support decisions about eligibility and priority for domestic-violence interventions in custody.

The published model includes predictors such as age, alcohol and other drug problems, criminal history, prior domestic-violence convictions, breaches of protection orders and Indigenous status.

Why it is concerning

A statistical risk score can influence who receives intervention, supervision or attention. The inclusion of characteristics that correlate with historical policing, disadvantage or over-representation raises questions about whether existing system inequalities can be reproduced through automated risk estimates.

Read more: www.correctiveservices.dcj.nsw.gov.au/resources/research-and-reports/corrections-research-evaluation-and-statistics/research-bulletin.html

DV-TRAS research bulletin: www.dcj.nsw.gov.au/content/dam/dcj/corrective-services-nsw/documents/research-and-statistics/DV-TRAS.pdf

Complain or raise a concern: NSW Department of Communities and Justice complaints: www.dcj.nsw.gov.au/contact-us/complaints-and-feedback.html

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

Corrective Services NSW has developed a broader family of automated and statistical risk tools, including the Custody Triage Risk Assessment Scale, Community Triage Risk Assessment Scale, automated assessment of sexual recidivism risk and a newer Triage Risk Assessment Scale for Sexual Offending.

These systems are designed to estimate future offending or return-to-custody risk and can inform correctional case management and intervention planning.

Why it is concerning

Predictive risk tools convert past administrative records and personal characteristics into estimates of future behaviour. Errors can affect intervention intensity, case planning and how a person is perceived by professionals. They also create a risk that historical patterns in policing and corrections become embedded in future risk classification.

Read more: www.correctiveservices.dcj.nsw.gov.au/resources/research-and-reports/corrections-research-evaluation-and-statistics/research-bulletin.html

Complain or raise a concern: NSW Department of Communities and Justice complaints: www.dcj.nsw.gov.au/contact-us/complaints-and-feedback.html

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

The NDIS Quality and Safeguards Commission uses large language models to annotate material including complaints, reportable incidents and audit reports. The system can identify risk indicators, classify themes and extract information about providers and workers. Those annotations can then be surfaced to regulatory officers.

The Commission states that the AI does not make the regulatory decision itself.

Why it is concerning

AI can influence which material appears significant to an investigator before the human decision is made. False summaries, missing context or incorrect risk labels could shape regulatory attention even where a person remains formally "in the loop".

Read more: www.ndiscommission.gov.au/about-us/corporate-reports/ai-statement

Complain or raise a concern: Complaints about the NDIS Commission: www.ndiscommission.gov.au/complaints/complaints-about-ndis-commission

Privacy complaints to the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

Changes to the NDIS Act allow greater use of automated administrative actions where government considers the relevant rules sufficiently clear and objective. The government states that human oversight and review rights remain, and that matters requiring judgement should be referred to a human decision-maker.

Why it is concerning

Automation in disability support can affect high-impact decisions at scale. Key issues include transparency about the rules being automated, whether individual circumstances can be adequately considered, how errors are identified and whether participants know when automation has contributed to an outcome.

Read more: www.health.gov.au/our-work/ndis-legislation-changes/amendments/ndis-amendment-securing-the-ndis-for-future-generations-bill-2026/about-the-changes-to-the-ndis

Challenge a decision: NDIS decision reviews: www.ndis.gov.au/participants/changing-your-plan/decision-reviews/how-request-review-decision

Services Australia uses supervised machine learning to identify Australian Government Disaster Recovery Payment claims that may warrant additional fraud checking. Flagged claims are stopped from automatically proceeding and referred to a human fraud analyst.

Why it is concerning

Even where a human makes the final decision, an algorithm can determine who receives additional scrutiny or delay. That raises questions about false positives, training data, whether particular groups are disproportionately flagged and how easily a person can understand why their claim was diverted.

Read more: www.servicesaustralia.gov.au/how-we-use-artificial-intelligence?context=22

Complain or raise a concern: Commonwealth Ombudsman, Services Australia complaints: www.ombudsman.gov.au/complaints/australian-government-agency-complaints/services-australia

Privacy complaints to the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

Revenue NSW uses analytics and machine learning in debt and fine administration. Models can contribute to identifying vulnerability, prioritising debt activity and determining whether particular enforcement pathways should be avoided or pursued.

Why it is concerning

Incorrect classification can affect how a person experiencing hardship is treated by government. A system designed partly to identify vulnerability can still cause harm if information is incomplete, inaccurate or interpreted without sufficient human context.

Read more: www.audit.nsw.gov.au/our-work/reports/revenue-nsw%E2%80%99s-administration-of-hardship-assistance

Complain or raise a concern: NSW Ombudsman: www.ombo.nsw.gov.au/

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

NSW social housing allocation uses an algorithmic process that places applicants into need categories and rank levels and helps determine which household is selected when a property becomes available.

Why it is concerning

Housing allocation involves extremely scarce resources and highly vulnerable applicants. Errors in category assignment, incomplete records or opaque prioritisation rules can materially affect how long a person waits for housing.

Read more: www.audit.nsw.gov.au/our-work/reports/social-housing

Complain or raise a concern: NSW Department of Communities and Justice complaints: www.dcj.nsw.gov.au/contact-us/complaints-and-feedback.html

NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

15 October 2026 update: The government announced an independent expert review of the Integrated Assessment Tool and its classification and prioritisation algorithms. Review announcement. Support at Home algorithm technical details, published 6 October.

The My Aged Care Integrated Assessment Tool uses structured information entered by human assessors and applies rules to support eligibility and classification decisions.

Why it is concerning

When assessment information is converted into an automated classification, people can be affected by how rigidly rules interpret complex health, disability, family and support circumstances. Transparency, review rights and the ability to correct inaccurate inputs are particularly important.

Read more: www.health.gov.au/our-work/single-assessment-system/about/tools

Integrated Assessment Tool user guide: www.health.gov.au/resources/publications/my-aged-care-integrated-assessment-tool-iat-user-guide

Complain or raise a concern: My Aged Care: www.myagedcare.gov.au/contact-us

Home Affairs uses AI across areas including service delivery, risk management, compliance and fraud detection, law enforcement and security, policy functions and administrative operations. AI can process personal information in these contexts.

Why it is concerning

Migration and border decisions can have major consequences for individuals while many risk models are not publicly transparent. People may not know which data points or inferred risk factors contributed to additional scrutiny.

Read more: www.homeaffairs.gov.au/commitments/Pages/Plans-policies-charters/ai-transparency-statement.aspx

Home Affairs privacy information: www.homeaffairs.gov.au/access-and-accountability/our-commitments/privacy

Complain or raise a concern: Home Affairs complaints: www.homeaffairs.gov.au/help-and-support/departmental-forms/online-forms/complaints-compliments-and-suggestions

Commonwealth Ombudsman: www.ombudsman.gov.au/complaints/how-to-make-a-complaint

SmartGates use facial recognition and electronic passport information to automate parts of immigration clearance at Australian borders.

Why it is concerning

Biometric identification has become routine infrastructure at the border. Issues include biometric accuracy, data retention, what happens when matching fails and the limited practical ability of travellers to avoid biometric processing.

Read more: www.immi.homeaffairs.gov.au/entering-and-leaving-australia/entering-australia/crossing-the-border

Complain or raise a concern: Home Affairs complaints: www.homeaffairs.gov.au/help-and-support/departmental-forms/online-forms/complaints-compliments-and-suggestions

Privacy complaints to the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

Queensland uses AI to analyse images from mobile phone and seatbelt enforcement cameras and identify potential offences for human review. The Queensland Audit Office reported around 208.4 million AI assessments during 2024.

Why it is concerning

These systems operate at population scale and capture large volumes of imagery involving drivers and passengers. The Queensland Audit Office identified risks involving privacy, inaccurate image recognition, handling and storage of photographs and the adequacy of human oversight.

Read more: www.qao.qld.gov.au/reports-resources/reports-parliament/managing-ethical-risks-artificial-intelligence

Complain or raise a concern: Office of the Information Commissioner Queensland: www.oic.qld.gov.au/community/how-to-make-a-privacy-complaint

Queensland Ombudsman: www.ombudsman.qld.gov.au/how-to-complain

NSW uses AI software to scan very large numbers of road-camera images and identify possible mobile phone and seatbelt offences. Potential offences then pass through multiple stages of human review before an infringement is issued.

The NSW Audit Office reported in 2026 that the seatbelt enforcement process uses AI elements to recognise patterns in photographs and make probabilistic classifications. The system is intended to scan more than 100 million trips and reach close to the entire NSW driving population each year.

Why it is concerning

This is population-scale automated visual surveillance. Although human review is required and privacy controls are built into the process, the initial identification of suspected offending is performed by proprietary AI at enormous scale.

Read more: www.audit.nsw.gov.au/our-work/reports/internal-controls-and-governance-2026-grants-consultants-purchasing-cards-and-technology

NSW Data Strategy example: www.data.nsw.gov.au/nsw-data-strategy/enabling-nsw-government-priorities

Complain or raise a concern: NSW Information and Privacy Commission: www.ipc.nsw.gov.au/privacy/citizens/make-complaint

Victoria uses AI-enabled cameras to photograph passing vehicles and automatically review the images for possible illegal portable-device use or seatbelt offences. Images flagged by the system are reviewed by qualified officers, and Victoria Police decides whether a fine is issued.

The images can show the faces of drivers and passengers, although the system does not use facial recognition to identify them.

Why it is concerning

The system conducts automated review of people inside vehicles on a large scale. Concerns include privacy, false classifications, retention of flagged imagery and the expansion of computer-vision enforcement into additional road-safety applications.

Read more: www.vic.gov.au/distracted-driver-and-seatbelt-detection-cameras

Complain or raise a concern: Office of the Victorian Information Commissioner: www.ovic.vic.gov.au/privacy/for-the-public/privacy-complaints/

The Australian Taxation Office uses AI and machine-learning models to identify possible non-compliance and support decisions about which matters warrant additional scrutiny.

An Australian National Audit Office review found dozens of AI models in production and identified shortcomings in governance, including incomplete data-ethics assessments for many models at the time of the audit.

Why it is concerning

Tax-risk models can influence who is investigated or contacted. The concern is not only whether the final decision is human, but whether an opaque risk score determines whose affairs receive attention.

Read more: www.anao.gov.au/work/performance-audit/governance-of-artificial-intelligence-the-australian-taxation-office

Complain or raise a concern: Tax Ombudsman: www.igt.gov.au/contact-us/

Privacy complaints to the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

AUSTRAC uses AI-enabled analytics across large volumes of financial transaction information to identify indicators of money laundering, terrorism financing and other financial crime and to produce intelligence for investigators and partner agencies.

Why it is concerning

Financial-intelligence systems operate across huge datasets and can infer suspicious patterns or relationships that individuals cannot easily inspect or challenge. AUSTRAC's intelligence function also sits within the National Intelligence Community, where transparency requirements differ from those applying to ordinary Commonwealth agencies.

Read more: www.austrac.gov.au/about-us/policies-and-governance/reports-and-accountability/our-ai-transparency-statement

Complain or raise a concern: AUSTRAC contact information: www.austrac.gov.au/contact-us

Privacy complaints to the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

ASIC uses AI in regulatory, surveillance, fraud and enforcement work. ASIC's AI transparency statement says there may be circumstances where it does not disclose use of AI in surveillance or enforcement activity.

Why it is concerning

That creates a transparency problem: the existence of an agency AI statement does not necessarily mean the public can identify every AI system affecting investigations or enforcement.

Read more: www.asic.gov.au/about-asic/what-we-do/how-we-operate/accountability-and-reporting/artificial-intelligence-transparency-statement/

Complain or raise a concern: Commonwealth Ombudsman: www.ombudsman.gov.au/complaints/how-to-make-a-complaint

Privacy complaints to the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

The Australian Public Service Commission has issued guidance acknowledging increasing use of AI in recruitment, including assessment of written applications, resumes, interviews and work samples.

Why it is concerning

Recruitment AI can reproduce discrimination present in historical data, misinterpret disability or communication differences and make it difficult for applicants to understand why they were screened out.

Read more: www.apsc.gov.au/working-aps/hr-practitioners/recruitment-aps/ai-recruitment

Complain or raise a concern: Raise concerns first with the recruiting agency. Privacy concerns can also be lodged with the OAIC: www.oaic.gov.au/privacy/privacy-complaints/lodge-a-privacy-complaint-with-us

Queensland Government staff use QChat, an approved generative AI tool. The Queensland Audit Office reported extensive use and identified risks including staff uploading protected information, relying on inaccurate output and using AI inappropriately to inform decisions.

Why it is concerning

A government-approved chatbot can become embedded in everyday administrative work before the public knows which decisions or documents have been influenced by it. Good governance depends on strong controls over sensitive information, verification and decision use.

Read more: www.qao.qld.gov.au/reports-resources/reports-parliament/managing-ethical-risks-artificial-intelligence

Complain or raise a concern: Queensland Ombudsman: www.ombudsman.qld.gov.au/how-to-complain

Office of the Information Commissioner Queensland: www.oic.qld.gov.au/community/how-to-make-a-privacy-complaint

Some Commonwealth consultation processes now disclose that approved AI tools may be used to identify themes, common issues and trends or assist with summaries of submissions. Human officers remain responsible for the final analysis.

Why it is concerning

Public submissions are intended to communicate citizens' views directly to government. AI summarisation creates another interpretive layer between the original submission and the official considering it. Minority views, unusual arguments and context can be lost when thousands of submissions are clustered or summarised.

Read more: www.pmc.gov.au/domestic-policy/office-ai/ai-infrastructure-consultation

Get involved: Where a consultation is open, submit directly and make the key points clear in the body of the submission rather than relying only on attachments.

The Australian Signals Directorate uses AI to help triage large volumes of data, identify potentially valuable intelligence and detect anomalous cyber activity.

Why it is concerning

Intelligence uses of AI operate with limited public visibility because operational secrecy is built into the system. This makes independent scrutiny of datasets, error rates, targeting criteria and downstream consequences difficult.

Read more: www.asd.gov.au/about/accountability-governance/ethical-ai-asd

Complain or raise a concern: Inspector-General of Intelligence and Security: www.igis.gov.au/making-complaint

Australian Defence is testing and developing AI-enabled military capabilities through AUKUS and domestic programs. Publicly documented examples include AI-assisted analysis of submarine information, autonomous drone swarms, AI-enabled object detection and tracking and autonomous maritime systems.

Why it is concerning

The closer AI moves towards target identification and military action, the more important questions of human control, error, escalation, accountability and the acceptable role of autonomous systems become.

Read more: www.defence.gov.au/defence-activities/programs-initiatives/aukus-advanced-capabilities/capability-achievements

AUKUS drone swarming: www.defence.gov.au/news-events/news/2026-06-22/aukus-partners-test-drone-swarming-tactics

AI-enabled targeting and reconnaissance: www.defence.gov.au/news-events/news/2026-07-08/new-drone-uses-ai-southern-jackaroo

Complain or raise a concern: Commonwealth Ombudsman: www.ombudsman.gov.au/complaints/how-to-make-a-complaint

For intelligence-agency conduct: www.igis.gov.au/making-complaint