The War on Terror and inception of Palantir
After 9/11, U.S. national security agencies pursued a sweeping “collect it all” surveillance program, turning to CIA-backed Palantir to manage and make sense of the troves of new data. Palantir’s software generated terrorist suspects and dictated investigators’ focus, tracking every movement, prayer, and sleep patterns of subjects. In a symbiotic relationship, investigators would then feed discovered data back into Palantir. For some of the 432,000 civilians who died during the war on terror, Palantir’s algorithm was the first link in the kill chain. From these beginnings, AI is now present in every stage of military decision making.
Israel’s use of Lavender AI
Israel’s Lavender AI system identified 37,000 people as potential bombing targets in the first weeks of the Gaza war. With 83% of Gaza’s war dead being civilians, the system was, according to an independent UN commission, committing AI-facilitated war crimes. This lethality is underpinned by a surveillance architecture that is total: facial recognition cameras blanket Palestinian cities, all calls and texts are intercepted, social media is monitored, and data is shared with the Israeli Defence Force (IDF) by US intelligence agencies.
Soldiers using Lavender AI were given 20 seconds to review each strike before approving it. One IDF soldier described the process as having “zero added-value as a human, apart from being a stamp of approval.” To accelerate the process further, the IDF pre-authorised civilian casualties based on Lavender’s target classifications, with 20 civilian casualties deemed acceptable for low-ranking targets.
The IDF uses Lavender AI to track targets back to their homes, deliberately targeting people when they are with their families. One IDF soldier noted that “It’s much easier to bomb a family’s home. The system is built to look for them in these situations.” This automated program was named “Where’s Daddy?.”
These AI targeted strikes violate two foundational principles of international humanitarian law: the obligation to distinguish between combatants and civilians, and the requirement that any civilian deaths be proportionate to concrete military advantage. Bombing 20 civilians to kill a low-ranking suspected militant satisfies neither. These AI-based targeting systems have relied on the technical support of US big tech companies such as Microsoft, Amazon and Google.
Speed over accuracy
AI targeting systems not only select who is killed, but where. A US strike on an Iranian girls’ school in Minab, Iran killed 168 people after an outdated targeting database failed to register that an erstwhile Islamic Revolutionary Guard compound had been repurposed as a school. The AI system had no context, no real-time human intelligence, and no mechanism to catch its own error.
These examples speak to the way militaries are prioritising the speed at which AI systems can operate, over the accuracy and careful consideration of each strike. The kill chain “find, track, target, strike” as demonstrated to journalist Bruno Maçães took 2-3 minutes from start to finish. Maceas was touring Palantir’s headquarters and was shown this process in relation to the war in Ukraine. Palantir’s CEO Alex Karp has claimed that advanced algorithmic warfare systems now carry the destructive asymmetry of tactical nuclear weapons. The structure of these systems rewards speed over deliberation, creating the conditions for a large number of catastrophic errors.
Much of the dialogue around military use of AI in warfare focuses on questions of human oversight. However, this assumes that AI has a legitimate role in making decisions that can end thousands of lives. But these systems are consistently faulty, repeat the biases in their training data, and cannot be properly checked.
Could Australia’s military use AI targeting?
Australia’s defence AI policy requires systems to address known biases, prioritise harm mitigation, remain explainable, and comply with domestic and international law. This means systems like Lavender AI should be ruled out.
However, the policy lacks any implementation roadmap or accountability mechanisms. When questioned by Senator David Shoebridge, The Department of Defence confirmed it uses Palantir products to “select targets on the battlefield.” The Department of Defence also confirmed it uses Maven - the same suite of Palantir products used by Israel to identify targets in Gaza, and that the US used in its targeting of the Minab girls school.
AI: trained for war, used on civilians
Palantir and ICE
In the US, Palantir built software for Immigration, Customs, and Enforcement (ICE) to track immigrants. The software gives agents access to a dataset of 20 million people, searchable by hundreds of highly-personal characteristics such as tattoos, scars, hair colour, eye colour, and license plate data. The US government has sought to further centralise government databases into Palantir’s platform, eroding the distinction between public and private databases and who has access to this highly sensitive and personal information. Under the ESTA data-sharing arrangement, Australians who have visited the US are likely included in this system.
Palantir in Australia
Palantir has a significant footprint in Australia too. At the federal level, Palantir has held contracts with The Australian Criminal Intelligence Commission, Department of Defence, Australian Taxation Office, Australian Transaction Reports and Analysis Centre, and the Australian Signals Directorate. Palantir has also partnered with Westpac, Rio Tinto, Westrac, and Coles, the latter of which uses the software to cut labour costs. We have little visibility into other commercial partnerships between Australian companies and Palantir.
What can we do?
Military-grade AI surveillance software has no place being commercially-available for use on unconsenting Australians going about their daily lives. Yet Australia’s privacy laws remain inadequate to the task. In 2023, the Albanese government committed to implementing a hundred privacy reforms but have only acted on a handful. Mandatory Privacy Impact Assessments (PIA) for high privacy risk activities, a right to delete data, and a fair-and-reasonable test for data-handling practices would each create meaningful accountability for these kinds of AI systems. None are yet in force.
While it is the responsibility of this government to fulfil the privacy promises they have made, Australians are not powerless in this fight. Consumers can choose not to shop with companies that deploy military-grade surveillance software. You can sign the petition at GetUp to demand Coles end their partnership with Palantir, and stay across privacy reform progress by signing-up for our newsletter and our privacy reform campaign.