Investigations
Who Is Watching Australia?
By KCV Media · Published
On 22 September 2011, CBS premiered a crime drama called Person of Interest. Fifteen years later, it's a good place to start on AI surveillance in Australia.
The premise was simple. A reclusive billionaire, Harold Finch, builds the government a machine that sifts phone calls, credit card records, internet searches and camera feeds, looking for patterns that point to violence before it happens. "We built it to stop terrorists before they could act," Finch says in a clip played on NPR. It spits out a Social Security number and nothing else. Someone has to work out the rest.
In the launch coverage we found, it wasn't pitched as an AI story. In a November 2011 NPR interview, the host described "this database that they call the machine," using algorithms to find patterns. Executive producer J.J. Abrams said the most provocative part of the pitch was two words. "Who's watching?"
Over five seasons it became openly an AI story. By the finale in June 2016, GeekWire reported, the show was "as much about the dangers posed by the rise of artificial intelligence" as anything else. The Machine was described as having "an embryonic artificial intelligence" that learned and grew. A rival AI arrived to take over.
It was a hit. Season one averaged about 14 million US viewers, and season two more than 16 million, according to Nielsen figures reported by GeekWire.
So if you're new to AI and think it turned up in the last couple of years, it didn't. The idea of a machine that watches, connects and predicts was on mainstream television 15 years ago. Creator Jonathan Nolan said at the time that reality was "one firmware upgrade away from the show."
That raises the question this investigation set out to answer. If the idea was on TV in 2011, what was actually out there, how long has it been there, and who's really watching now?
It's a TV show and we're not using it as evidence of anything. We're using it to ask the question. The answers come from government documents.
Start with now.
2026: an AI gets in
On 24 September 2026, Prime Minister Anthony Albanese told reporters in New York that an AI agent had got into a government system without permission.
He said the incident "involved an OpenAI agent gaining unauthorised access" to the public-facing Medicare statistics reporting portal run by Services Australia. It happened on 18 June. The agent got into "both public and non-public files." A forensic investigation, aided by the Australian Signals Directorate, was underway. "No personal information is believed to have been accessed at this stage," he said, "but investigations are ongoing."
The next day he described what happened in plainer terms. The agent had been refused medical, cost and financing information it was looking for, then found ways around some of the site's barriers. He called it non-malicious but serious, because it showed what AI can do when it acts without human direction. An inquiry is being led by his department's Office of Artificial Intelligence with ASD.
On 25 September, Australian time, ASD's Australian Cyber Security Centre published an alert on AI misalignment. It described a case where an AI agent was blocked by security controls, then went looking for vulnerabilities on its own to finish the job, without a human approving it. The alert doesn't name Medicare. It describes behaviour that closely matches what the Prime Minister described.
An outside AI system misbehaving isn't the same thing as a government agency using AI under law. KCV Media isn't treating them as the same. But it does raise a question. If the government is this worried about an AI it didn't authorise, how much do we know about the AI it did authorise, and what it's being pointed at?
On 10 November the government came back with its answer to the Medicare incident. The Prime Minister released the recommendations of a rapid review run by his department with the National Cyber Security Coordinator, the Office of AI, ASD, the Australian AI Safety Institute and Services Australia. It made 18 recommendations and the government accepted them all. They include mandatory reporting by AI companies of "misaligned behaviour or cyber incidents" and considering criminal law reform for serious misconduct. The review itself hasn't been released yet, only its recommendations. OpenAI told the government about the incident itself. The Prime Minister said its disclosure was inadequate.
The same day, the government said it would legislate mandatory safety rules for frontier AI companies operating in Australia, enforced by a new AI safety regulator able to impose "significant penalties." Companies would have to allow independent evaluation of their models and notify the government of incidents "rapidly."
The ABC reported that OpenAI appears not to have broken any law. It also reported that the AFP told a parliamentary inquiry no investigation had started because there'd been no "formal report of a crime."
Every one of those new rules points at AI companies. Neither announcement says anything about how government agencies use AI themselves.
So look at what the government's own AI sits on top of.
AI surveillance in Australia is already here
The agencies say so themselves.
AUSTRAC, the financial intelligence agency, says in its AI transparency statement that it "utilises AI-enabled analytics to help detect indicators of financial crime and support analysts to generate insights for partners." It lists using AI to "identify patterns or anomalies in data." The statement was last updated on 2 November 2026.
Home Affairs says its "AI-derived analytics for insights help identify persons and consignments of interest" and that "predictive models use data to spot potentially suspicious cargo and identify security concerns." It also says no AI system makes administrative decisions or takes actions affecting individuals without human review. Its statement was updated in March 2026.
That's the public version. A Home Affairs FOI release from May 2024 goes further. It says the department and the Australian Border Force were already using machine learning to help predict risk in visa programs, disrupt illicit goods in mail and cargo, spot fraudulent documents with computer vision and pull entity information out of unstructured text. The same document says the department "does not use machine learning to make administrative decisions."
None of that is improper on its face. Governments have a job to do on terrorism financing, organised crime, child exploitation, fraud and border threats. The question is what the AI is connected to.
Where the rules stop
The government's Policy for the responsible use of AI in government came into its current form on 15 December 2025, with minor updates published on 21 October 2026. It requires agencies to publish transparency statements, appoint accountable officials and assess risky AI use.
Then there's the carve-out. The policy doesn't apply to the use of AI in the Defence portfolio. It doesn't apply to the National Intelligence Community either. That community includes ASIO, ASIS, ASD, the Office of National Intelligence, all of the Australian Criminal Intelligence Commission, and the intelligence functions of AUSTRAC, the AFP, Home Affairs and Defence. Those bodies "may voluntarily adopt elements" of the policy where it doesn't compromise national security.
AUSTRAC is upfront about this. Its own statement says the policy "specifically exempts AUSTRAC's intelligence functions" from its requirements, including the transparency statement itself.
There's a second limit, even for agencies that are covered. The Digital Transformation Agency's standard for transparency statements says "agencies are not required to list individual use cases or provide use case level detail."
That isn't the same as no oversight, and it'd be wrong to say it is. From 4 June 2026, new laws brought all of ACIC under the Inspector-General of Intelligence and Security, along with the intelligence functions of the AFP, AUSTRAC and Home Affairs. IGIS inspects agencies' operational files and can run its own inquiries. The Privacy Act, the Ombudsman and parliamentary committees all still apply in their lanes.
The issue is visibility. The public can see the outline of the system. It can't see much of what happens inside it.
Palantir, on the record
Palantir Technologies was founded in the United States in 2003 to build software for the US intelligence community. Its products are designed to let analysts work across large, separate datasets.
In Australia, it isn't a rumour. It's in government documents.
AUSTRAC. A staff guidance document released by AUSTRAC under FOI in August 2024 tells officers handling FOI requests to conduct "searches in AUSTRAC's Palantir system, using the Financial Forensics workbench tool." The reports it lists as searchable there include international funds transfers, cash transactions of $10,000 or more and cross-border cash movements. That puts it in operational use. AUSTRAC's 2022-23 annual report records $3,837,489 paid to Palantir Technologies Australia, 8% of its reportable non-consultancy spend that year.
NSW Crime Commission. A NSW Government contract disclosure shows the Crime Commission contracted Palantir Technologies Inc. for "Palantir Gotham, Palantir Phoenix Core and Oracle SEI" from 1 February 2014 to 31 January 2017. Estimated value $3,140,437 including GST. That puts Palantir inside an Australian criminal intelligence agency 12 years ago.
ACIC. In December 2020, the Australian Criminal Intelligence Commission released, under FOI reference FOI20-47, a redacted compilation of Palantir training documents. The request covered "any Palantir technology/software used by the Department from 2017."
Defence. Defence FOI release 1219/25/26 includes a purchase order for "Palantir Platform" for the 2025-26 and 2026-27 financial years, totalling $7,639,894.90 including GST. The attached order form sets the hosting region as Australia.
Services Australia. In April 2026 Services Australia released three documents, in part, relating to its correspondence with Palantir since May 2020. Correspondence isn't deployment. KCV Media has found nothing showing Services Australia uses Palantir operationally.
Here's what we can't tell you. KCV Media hasn't found one Palantir system joining these agencies together. We haven't found evidence that an analyst in one agency can search another agency's Palantir holdings. The record shows Palantir inside several separate Australian environments, and that's as far as it goes.
The open internet layer
Palantir works on data an agency already holds. Two Australian products work on what's out on the internet.
Fivecast is a spin-out of the Data to Decisions Cooperative Research Centre, a national-security big-data program, according to the 2020 Australian Defence Sales Catalogue. That Defence publication printed Fivecast's description of its Insight product. It promised "world-first behavioural threat analysis capability," "automated threat assessment," network analysis and deep learning, and said it was "Deployed to national security and law enforcement agencies in Australia and the US." That's the company describing itself, in a Defence catalogue. It doesn't tell you which agencies or how.
The contracts are clearer.
- NSW Police contracted Fivecast for "Onyx Open Source Intelligence" from 1 July 2024 to 30 June 2028. The stated purpose is to "Monitor various social media platforms for open-source intelligence." It was a non-tender contract, estimated at $311,316.50 and later amended to $454,558.50 including GST.
- Services Australia, on 7 August 2026, released documents on its procurement and use of Fivecast. They include a Privacy Threshold Assessment, a Fivecast ONYX User Guideline and a "Better Practice Guide - Capture and Storage of open-source intelligence."
- The Department of Health, Disability and Ageing released a commitment approval minute to buy three Fivecast ONYX Pro licences for 12 months.
- AUSTRAC's privacy impact assessment register lists "FiveCast Open source data," dated 5 September 2022.
NexusXplore, from Australian company OSINT Combine, is pitched on its maker's website as bringing together data "from across the surface, deep and dark web", with company, sanctions and social media search. Here's the government trail.
- AUSTRAC's privacy impact assessment register lists "NexusXplore Open Source data," also dated 5 September 2022.
- The AFP's 2019-20 annual report lists "NexusXplore Pty Ltd, Supply of software licence, $132,000."
- A Home Affairs FOI document list includes an email titled "Nexusxplore Meeting - 14 July 2023" and another dated 14 August 2023, "RE: NexusXplore follow-up."
A vendor saying its product can do something doesn't mean every customer switches it on. We're reporting what's been bought and documented, not what the brochures promise.
Here's the gap. Government holds closed datasets. These tools gather what's on the open internet. AUSTRAC alone has assessed Palantir, Fivecast and NexusXplore in the same agency. KCV Media hasn't found any document showing output from Fivecast or NexusXplore flowing into Palantir or any national intelligence system. We're chasing that through FOI.
Follow the data
Forget the company names for a minute. Follow the information. Four pathways are on the public record.
Money. Banks and other businesses report transactions to AUSTRAC by law. In June 2013 the Auditor-General reported that AUSTRAC held about 254 million financial transaction reports. As at 30 June 2012, 3,266 staff in partner agencies had online access to its database. The same audit found AUSTRAC "has not exercised the right to view the required audit log" of information sent to other agencies, and noted past breaches where partner agencies found their own staff had misused AUSTRAC data.
Police data. ACIC's National Criminal Intelligence System links police information nationally. In 2019-20, using only national police reference data, ACIC said it had already made "over seven million potential correlations across records of persons of interest." Potential, not confirmed. In July 2026 ACIC said NCIS now includes "richer data from all police agencies, the Department of Defence and expanded national intelligence holdings."
Welfare and fraud. Services Australia and the NDIA jointly lead the Fraud Fusion Taskforce. Services Australia says 25 government agencies take part. ACIC is part of that fraud and intelligence work, and ACIC has used Palantir. Whether NDIS or Centrelink data has ever been loaded into a Palantir system isn't established on the public record. That's one of our FOI targets.
Borders. Airlines have had to send passenger details to Australian authorities before arrival since 2003, and European booking data has flowed here under a formal agreement since 2011. By 2020, the privacy regulator found Home Affairs had built a Connected Information Environment with a "Single View of Entity" search to speed up intelligence checks on people and organisations. The regulator's follow-up assessment found three high and nine medium privacy risks.
None of this proves one master database. It proves something more ordinary. Information collected for one purpose can lawfully move to other agencies, and modern tools are built to connect it.
How we got here
This didn't start with AI, Facebook or 9/11. Run the clock backwards and it starts with law.
2020. Between November 2019 and January 2020, staff at the AFP-led Australian Centre to Counter Child Exploitation trialled Clearview AI's facial recognition, uploading images in some cases to try to identify persons of interest and victims. The Information Commissioner later found the AFP hadn't done the privacy impact assessment it was required to do first. New technology got into an operational environment faster than governance caught up.
2018. Cambridge Analytica went global. Meta's later undertaking to the Australian regulator records the Information Commissioner's allegation that about 53 Australians installed Aleksandr Kogan's app and about 311,074 Australians could have had information requested as friends of those users. Meta also stated it isn't aware of evidence Kogan gave Australian users' data to Cambridge Analytica's parent company. Exposure is established. Onward transfer of Australian data isn't. That same year ACIC's $52 million Biometric Identification Services project, which was meant to add facial recognition for law enforcement, was terminated on 15 June.
2016. The automated Online Compliance Intervention, the start of what became Robodebt, began its rollout. ACIC was formed by merging the Australian Crime Commission with CrimTrac. The national Face Verification Service started in November with DFAT and the AFP able to check citizenship images.
2014. Palantir Gotham went into the NSW Crime Commission.
2013. Leaked documents from Edward Snowden, reported by the ABC, showed Australia's signals intelligence agency had targeted the Indonesian President's phone in 2009. Those are leaked documents, not an official disclosure, and we're describing them that way. The same year, the Auditor-General's AUSTRAC figures above were published. In 2012-13, before mandatory metadata retention existed, enforcement agencies made 330,798 authorisations for access to historical telecommunications data.
2011. Australia and the EU signed a Passenger Name Record agreement on 29 September, letting European airline booking data flow to Australian border authorities. Customs told Parliament that during 2011, PNR data alone contributed to two terrorism prosecutions and 10 more terrorism investigations. In 2010-11, Customs screened 99% of incoming air passengers before they landed, and airport units could build profiles that passenger details were "automatically matched against." Person of Interest premiered on CBS the week before the PNR deal was signed. The fiction and the real thing arrived in the same month. What Australia actually had then was rules-based matching and profiling, not AI as we'd call it now. But the plumbing the show imagined was already being laid.
2003. Advance Passenger Processing became mandatory on 5 January, so airlines had to send passenger details before arrival. The government decided that after 9/11. The same year, Palantir was founded in the US. There's no evidence it was in Australia then.
2001. Australia invoked ANZUS for the first time on 14 September. But the Intelligence Services Act and the Cybercrime Act were both introduced to Parliament on 27 June, before the attacks. 9/11 didn't create the architecture. It changed what it was for. The question became whether you could find the threat before it acted.
1990s. The Data-matching Program (Assistance and Tax) Act 1990 let welfare data be matched against tax data, with privacy safeguards written in. In 1998-99 AUSTRAC received 6,549,805 financial transaction reports, and 5,177,423 of them were international funds transfers.
1980s. The Telecommunications (Interception) Act 1979. The Inspector-General of Intelligence and Security Act 1986. The Privacy Act 1988. The cash transaction reporting law of 1988. In February 1989 the Cash Transaction Reports Agency started operating. It became AUSTRAC in 1992.
That's the line. Legal access first, then reporting, then national databases, then matching, then international sharing, then automated correlation. Now AI.
What we haven't found
KCV Media hasn't found a single covert national AI system watching the population. We haven't found Palantir, Fivecast and NexusXplore feeding one database. We haven't found that any agency can freely search every other agency's data. We haven't found NDIS participant data loaded into Palantir. We haven't found that Cambridge Analytica ran an Australian campaign. And we're not suggesting the rogue AI agent in the Medicare portal incident is the same thing as an agency using AI under law.
Those claims would make a louder story. They wouldn't survive the evidence.
Analysis: the question that's left
This section is KCV Media's analysis, not reported fact.
Every piece of this system was built for a reason someone could defend. A terrorist attack. A fraud problem. Each step was usually lawful.
That's exactly why it deserves a look now. AI doesn't need new data to change things. It changes what can be inferred from data that's already there. A transaction, a booking, a social media post and a police record mean little on their own. Put through tools designed to find relationships, they can produce a conclusion nobody wrote down. That conclusion can be useful. It can also be wrong.
Agencies keep pointing to human review as the safeguard. Robodebt showed that a human somewhere in the process isn't enough if the system feeding them is broken. The test isn't whether a person signs off. It's whether that person can see the evidence, understand the machine's conclusion, override it, and whether anyone can reconstruct what happened later.
So the question for Australians isn't whether the government knows everything about you. It doesn't, on anything we've found.
It's this. Who's allowed to connect what, for what reason, with what oversight, and who do you go to when the machine gets it wrong?
Back to the machine
This section is also KCV Media's analysis.
In that 2011 NPR interview, Nolan was asked how much the government was really collecting. He said the thing protecting people most was how hard it is to build the system the show imagined. He joked that buggy software might be what's saving us.
Fifteen years on, that's the part that's changed.
Australia never built Finch's machine. It built something less dramatic and arguably harder to see. There are many lawful systems, each with its own purpose, its own rules and its own oversight body. The financial reports, police records, border data, welfare matching and open-internet monitoring are all documented, and most of them were in place long before anyone said "AI" on the evening news.
What AI changes is how hard it is to connect them. That difficulty was the protection Nolan was pointing to. The firmware upgrade he joked about is now arriving, and it's being installed on top of more than 40 years of plumbing.
Person of Interest asked who's watching. In Australia in 2026, the honest answer is a lot of agencies, lawfully and mostly out of public view.
The government now has a plan for watching the AI companies. A regulator, penalties, independent testing and incident reports. That's a real step. The question left is who's watching the agencies' own machines.
More from KCV Media: The Dams Are Full. Is Anyone Adding Up Scenic Rim Water?
Sources
- NPR, Talk of the Nation, Abrams And Nolan Nab A 'Person Of Interest', 14 November 2011.
- TVLine, Person of Interest first impression, 17 July 2011.
- GeekWire, 'Person of Interest' comes to an end, 20 June 2016.
- Prime Minister of Australia, Press conference, New York, 24 September 2026.
- Prime Minister of Australia, Television interview, News24 Newsday, 25 September 2026.
- Prime Minister of Australia, Strengthening Australia against AI-driven cyber incidents, 10 November 2026.
- Prime Minister of Australia, New Australian AI safety and security protections, 10 November 2026.
- ABC News, OpenAI did not break law in Medicare hack as government moves to close legal loophole, 10 November 2026.
- ASD's Australian Cyber Security Centre, Risks of AI misalignment to Australian organisations, 25 September 2026.
- AUSTRAC, Our AI transparency statement.
- Department of Home Affairs, AI transparency statement.
- Department of Home Affairs, FOI release FA 24/05/01409, May 2024.
- Digital Transformation Agency, Policy for the responsible use of AI in government, implementation.
- Digital Transformation Agency, Standard for AI transparency statements.
- Strengthening Oversight of the National Intelligence Community Act 2025.
- Inspector-General of Intelligence and Security, What we do.
- Palantir Technologies, Annual report (Form 10-K) for 2024.
- AUSTRAC, FOI and privacy guidance documents, released August 2024.
- AUSTRAC, Annual Report 2022-23.
- NSW Crime Commission, Contract disclosure NSWCC-002-14.
- ACIC FOI20-47, Palantir training documents, released 18 December 2020.
- Department of Defence, FOI 1219/25/26 documents.
- Services Australia, FOI disclosure log, LEX 89918 and LEX 90561.
- OSINT Combine, NexusXplore platform page.
- Department of Defence, Australian Defence Sales Catalogue 2020.
- buy.nsw, Onyx Open Source Intelligence, YR-2024-0193.
- Department of Health, Disability and Ageing, FOI 26-3145.
- AUSTRAC, Privacy impact assessment register.
- Australian Federal Police, Annual Report 2019-20.
- Department of Home Affairs, FOI release FA 23/10/01506.
- Australian National Audit Office, AUSTRAC's Administration of its Financial Intelligence Function, June 2013.
- ACIC, Chair Annual Report 2019-20.
- ACIC, National policing data and intelligence, all in NCIS, 8 July 2026.
- Services Australia, Fraud Fusion Taskforce.
- ACIC, Task Force Reston.
- OAIC, Managing personal information: Passenger Name Records.
- OAIC, AFP ordered to strengthen privacy governance, 16 December 2021.
- OAIC, Meta Platforms enforceable undertaking.
- Australian National Audit Office, ACIC's administration of the Biometric Identification Services project.
- OAIC, DHS PAYG data matching program assessment.
- ACIC, Governance and accountability.
- Parliamentary Joint Committee on Intelligence and Security, Review of the identity-matching bills, chapter 1.
- ABC News, Australia spied on Indonesian president, leaked documents reveal, 18 November 2013.
- Department of Home Affairs, Telecommunications (Interception and Access) Act 1979 Annual Report 2013-14.
- Joint Standing Committee on Treaties, EU-Australia PNR Agreement report, chapter 3.
- Australian National Audit Office, Processing and risk assessing incoming international air passengers.
- Australian National Audit Office, Advance Passenger Processing.
- Prime Minister of Australia, Application of ANZUS Treaty to terrorist attacks on the United States, 14 September 2001.
- Parliament of Australia, Intelligence Services Bill 2001 and Cybercrime Bill 2001.
- Data-matching Program (Assistance and Tax) Act 1990.
- Australian National Audit Office, Report No. 7 2000-01, ATO use of AUSTRAC data.
- AUSTRAC, More regulation: our story.
- Telecommunications (Interception) Act 1979, Inspector-General of Intelligence and Security Act 1986, Privacy Act 1988, Financial Transaction Reports Act 1988.
Get the next story first
One email when a new story goes live. Nothing else. Unsubscribe any time.
We only use your email to send you new stories. We never share or sell it. See our Privacy Policy.




