.png)
The Watching Machine: The Privatisation of Surveillance Infrastructure
- THE GEOSTRATA
- 5 days ago
- 5 min read
Nobody voted for Palantir to run immigration enforcement. Nobody elected the NSO Group to decide which migrants get surveilled. And nobody asked Clearview AI to scrape over 60 billion publicly available images into a facial recognition database and then license it to immigration investigators operating in contexts where algorithmic error has no formal accountability mechanism.
Illustration by The Geostrata
Yet here we are. Across the United States, the European Union, and a growing number of governments in between, the apparatus of migration control has been quietly contracted out to private technology firms whose tools are more powerful, less accountable, and more legally ambiguous than almost any publicly debated policy. The infrastructure of who gets flagged, detained, and deported is now, in significant part, privately owned.
What does this look like in practice? What does it mean for the people it is used against? And how far does it reach before it collides with the body of international law designed to protect the most vulnerable people on earth?
ORIGINS AND WORKING OF THE STATE OF SURVEILLANCE CREATED FOR MIGRANTS The shift accumulated through contracts, which were hardly ever scrutinised. Palantir’s first ICE contract dates to 2011. Total ICE obligations to the company have since reached approximately $248 million. The most significant recent addition is a $30 million contract for ImmigrationOS, which is a platform designed to integrate data from federal, state, and commercial databases to track, identify, and apprehend individuals prioritised for removal.
ImmigrationOS does not operate alone.
Palantir’s tools sit alongside Clearview AI’s facial recognition, BI2’s iris scanners, and Paragon’s phone-extraction software. These platforms combine social media posts, travel records, tax data, and phone extractions into unified investigative files.
The result is not a file. It is a profile assembled without the subject’s knowledge, from data they never consented to share. The architecture is not uniquely American, as Palantir holds over £900 million in British public sector contracts and operates in at least three German federal states. Accountability does not transfer with the contracts.
WHAT HAPPENS WHEN THE ALGORITHM IS WRONG
Here is the part of this story that gets buried in technical language and should not be.
Facial recognition has shown disproportionately high error rates for people of colour, women, and children. Amazon’s Rekognition incorrectly matched the photos of 28 US congressmen with criminal faces. It has an error rate of up to 39% for non-Caucasian individuals. These are not edge cases. They are structural features of systems trained on datasets that overrepresent white male faces and underrepresent everyone else.
The consequences at the border are not abstract either. The EU’s iBorderCtrl programme, piloted in Hungary, Greece, and Latvia, used webcam-based lie detection that analysed facial muscle movements to flag deceptive answers. Because the algorithm was trained on white and emotionally neutral male faces, it systematically flagged trauma survivors, racial minorities, and women as deceptive.
For example, A Congolese asylum seeker who cannot control the involuntary muscle response triggered by a traumatic memory does not fail the test because they are lying. They fail because the training data did not include them.
The American CBP One application, used by asylum seekers to schedule border appointments, requires a facial liveness selfie. Civil rights groups have documented systematic rejection of darker-skinned applicants, yet CBP does not collect or disclose race or ethnicity in its performance data, leaving the systemic bias invisible to the statistics. The people being rejected have no appeal mechanism against an algorithm they cannot see, operated by a contractor who answers to no ombudsman.
The ACLU has documented that in New Orleans, nearly every use of facial recognition from October to August was conducted on a Black person. In Detroit, all 129 facial recognition searches in 2020 were conducted on images of Black people. The technology is not neutral. It never was. It reflects the racial architecture of the enforcement system it was built to serve.
THE USE OF PEGAUSUS AND THE PRIVATISATION OF TARGETED SURVEILLANCE
Pegasus was developed by Israel’s NSO Group. Pegasus infiltrates smartphones without interaction with the user. It is a no-click installation that grants full access to messages, emails, camera, microphone, and location data. This spyware leaves no digital trace of the intruder’s identity, and it offers high-intelligence returns with minimal risk of detection. This makes it a tool sought by all forms of governments. Its targets have ranged from suspected criminals to journalists, human rights lawyers, and political opponents.
A Harvard thesis published in 2025 documented Israel’s use of Palestine as a testing ground for Pegasus development before it was exported globally. The pattern is consistent, as the tools refined on occupied or surveilled populations are commercialised, exported, and eventually deployed on migrant and minority communities in countries that consider themselves democracies.
The leaked Pegasus target list contained over 50,000 phone numbers concentrated in countries like Azerbaijan, Bahrain, Hungary, etc. Many of these countries have significant records of using surveillance against migrants, refugees, and ethnic minorities. The NSO group has consistently maintained that Pegasus is sold only to vetted government clients for legitimate security purposes.
In May 2025, a US jury initially awarded $167 million in damages to Meta; a federal judge subsequently reduced that figure to $4 million while imposing a permanent injunction barring NSO from targeting WhatsApp. The court’s finding was unambiguous: the damage was real, the victims were real, and the company’s claim of government-client insulation did not hold.
THE LEGAL GAP IN THE INTERNATIONAL HUMANITARIAN LAW
This is where the working of private surveillance meets the architecture of international law; here, the collision is most consequential and most ignored. International Humanitarian Law is designed to protect civilian populations from disproportionate harm during conflict. But migration enforcement is not classified as armed conflict, and so its most aggressive tools operate in a legal grey zone that IHL cannot fully reach.
The instruments that can and should apply as the 1951 Refugee Convention, the International Covenant on Civil and Political Rights, and the UN Convention Against Torture prohibit refoulement, arbitrary detention, collective punishment, and treatment that degrades human dignity. The use of AI in immigration contexts has already resulted in wrongful detentions, misidentifications, and an erosion of due process that these instruments are designed to prevent.
The problem is enforcement. A country that uses a biased algorithm at a border checkpoint has not technically issued an expulsion order. It has allowed a private contractor’s system to generate a flag, which a human official can then act on. The chain of legal responsibility is deliberately obscured.
The increased privatisation and algorithmisation of national security through machine learning, AI, and spyware pose unprecedented risks to fundamental rights and create significant legal uncertainty about where international human rights obligations actually apply.
The EU AI Act came into force in 2025. This bans real-time facial recognition in public spaces and prohibits social scoring. Most of the world has no equivalent restrictions. The surveillance infrastructure being built today, along with the biometric databases, AI-powered cameras, mandatory digital ID, and integrated commercial spyware, will be extremely difficult to dismantle once embedded in enforcement bureaucracies that benefit from their existence.
There is a version of this story with a resolution. It involves binding international standards for algorithmic transparency in migration enforcement, mandatory publication of error rates disaggregated by race and ethnicity, a ban on commercial spyware against civilian populations, and the extension of existing human rights frameworks to private technology contractors operating in public enforcement contexts.
That version does not yet exist. What exists instead is a global market in surveillance tools, a patchwork of national laws that companies routinely route around, and a growing population of people whose fates are being decided by systems they cannot see, challenge, or appeal. In the architecture of migration control, the most consequential decisions are the ones being made by no one in particular and for which no one can be held accountable.
BY RAGHAV GUPTA
TEAM GEOSTRATA
Lately, I was searching for a reliable download ig reels tool that works without installing any application. After trying a few different websites, this one worked surprisingly well for me and the process felt very simple.z
The growing role of private companies in surveillance and migration enforcement raises serious questions about transparency and accountability. The examples of biased algorithms show how technology can affect real lives when oversight is limited. While reading, I reflected on how different experiences shape our sense of fairness. When I play basketball legends , a game that helps me relax, have fun, and feel good, I am reminded that equal treatment and trust are values that matter both in games and in society.
Geometry dash subzero is a high-speed platformer that challenges your reflexes, coordination, and focus from the very first second. A small cube that autonomously advances through a frozen, obstacle-filled world is under your control. There are no shortcuts, no control over pace, and no slowing down. The sole objective is to tap at the optimal moment to leap, fly, or flip your way to the finish line.
This article was so engaging that I immediately wanted to read more from you the moment I finished. You take a complicated topic and make it feel approachable without ever oversimplifying anything. I appreciate how every point is backed up with a clear example rather than left hanging. The structure makes even the trickier sections easy to follow from start to finish. I came away genuinely better informed than when I started. Thanks for such a thoughtful and well-researched post.
This cleared up so much of my confusion about the whole process. I appreciate that you focused on the practical realities, not just the looks. The points on support and fit made granite vanity countertop installation far less mysterious to me. It's obvious you know this thoroughly. Bookmarking this one for sure. Thank you.