Code, Contracts, and Complicity 2.0
How AI and Governments Are Closing Ranks on Dissent
First published in The Signal Dispatch, 25 August 2025. Republished here in revised and expanded form, with substantive updates throughout and one new section — “The Manifesto” — drafted in April 2026 in response to Palantir’s published 22-point doctrine.
The Day Democracy Showed Its Digital Face
On 9 August 2025, British democracy revealed what it had become. In Parliament Square — where suffragettes once demanded votes and millions marched against war — police systematically arrested 522 people for holding cardboard signs that read: “I oppose genocide. I support Palestine Action.” By day’s end, 532 had been arrested — the largest mass arrest in London since the 1960s.¹
This wasn’t disorder being contained; it was dissent being catalogued and criminalised. Officers drawn from forces across the country moved through the crowd with mechanical efficiency, processing arrests of peaceful protesters with the rhythm of an assembly line.
The mass arrests followed months of escalating repression documented by civil society groups. In May 2025, Bond — the UK network for international development organisations — warned that facial recognition technology was being deployed at peaceful gatherings, “violating privacy rights and deterring campaigners from participating in demonstrations.”² Their annual review found that UK police were arresting climate protesters at three times the global average rate, with some receiving five-year sentences merely for participating in protest-planning video calls.
Whether facial recognition was deployed that day remains unclear. But the operation unfolded against this documented backdrop of AI surveillance expansion, technology that Bond noted “disproportionately misidentifies people of colour, increasing the risk of wrongful arrest.”³ It is hard to believe the technology has not been used here given the history of this government. The infrastructure exists; the only question is when, not if, it will be turned on every protest.
The technology enabling this transformation wasn’t designed in some authoritarian backwater. It was built in Silicon Valley by companies that promised to “democratise AI” and “benefit humanity.” The same executives who speak at conferences about ethics and safety are selling artificial intelligence to militaries and police forces, teaching algorithms that their highest purpose is to identify, track, and neutralise human beings.
This is a story about betrayal — how the AI revolution we were promised became a counter-revolution against human freedom. It’s about how our governments and tech giants formed an unholy alliance, turning tools of liberation into instruments of oppression.
The Promise and the Betrayal
Remember the promises? AI would cure cancer, reverse climate change, unlock human creativity. Tech CEOs stood on stages and promised a better world.
Instead, they built the perfect surveillance state. The same algorithms meant to optimise traffic now optimise authoritarian control. The facial recognition that would help find missing children hunts those who dissent.
This isn’t technological determinism — it’s a choice made in boardrooms where quarterly earnings matter more than human lives, where executives know exactly what their systems enable but hide behind the language of “dual use.” The same technologies sold to us by their right hand to heal humanity are being unethically developed for surveillance, control, even kill chains in the left.
The Military-Industrial-AI Complex
The corruption begins with contracts worth billions, signed between tech giants and defence departments. These aren’t partnerships to protect democracy — they’re agreements to automate oppression.
The scale is no longer the kind of thing a democracy can plausibly oversee. American tech firms now hold defence and intelligence contracts worth tens of billions of dollars, awarded under classification regimes that keep most of the work hidden from the press, from Congress, and from the engineers building it. The same companies that ask for our trust to handle our email, our search history, our health records, and our children’s photographs are simultaneously the prime contractors for cloud platforms hosting military intelligence, AI models analysing battlefield data, and biometric systems feeding kill chains. There is no firewall between the consumer division and the defence division. There is one company, with one set of capabilities, sold to whoever can afford it.
The language used to describe this fusion — “dual use,” “public-private partnership,” “national security innovation” — was engineered to reassure. It functions instead as cover. “Dual use” means a technology built for surveillance can be marketed as crime prevention; a system built for war can be sold as disaster response; a tool built to track a population can be described as customer analytics. The framing collapses the moral distinction between civilian and military application, and in collapsing it, removes the basis on which the public might object. By the time a citizen learns that the company storing their cloud backup is also the company building the targeting system bombing a refugee camp, the contract has already been signed, the systems are already deployed, and the engineer who might have refused has already been replaced.
What follows is not an exhaustive accounting. It is a set of representative cases. Each shows the same pattern: a company that built its public reputation on benign tools is now embedded in the architecture of state violence, at home and abroad.
Microsoft: The Pentagon’s Primary Partner
Microsoft was awarded a US$22 billion contract in 2021 to provide “Integrated Visual Augmentation Systems” (IVAS) to the US military — AI-powered headsets that would turn soldiers into nodes in a vast killing machine.⁴ The programme was beset by years of failures: soldiers reported headaches, eyestrain, and nausea so severe that Congress threatened to cut funding. In February 2025, Microsoft announced it was transferring the contract to Anduril Industries, the defence-tech firm founded by Oculus VR creator Palmer Luckey. The Army formally novated the contract on 10 April 2025.⁵ Microsoft remains as the cloud provider; Anduril now controls production, hardware, and software.
The handover did not end Silicon Valley’s IVAS bet. It deepened it. Anduril is among the most aggressive military AI firms in the United States, founded explicitly to bring Silicon Valley’s “move fast and break things” ethos to weapons systems. The contract did not dissolve. It changed hands within the same ecosystem, to a more openly militarised actor. Strip away the corporate speak, and you still have AI-powered headsets that turn soldiers into nodes in a vast killing machine. Azure cloud services still host military data. AI models still analyse intelligence. Machine learning systems still identify targets. The names on the contract change. The architecture does not.
Microsoft’s reach extends well beyond the Pentagon. The company operates an Azure Israel cloud region serving government and public sector customers,⁶ and as we will see when we reach Gaza, Microsoft’s role there is now a matter of public record acknowledged by the company itself — not inference, not allegation, but documented fact.
Google: From “Don’t Be Evil” to “Don’t Get Caught”
Google’s transformation from idealistic startup to surveillance contractor is complete. After employee protests forced it to abandon Project Maven in 2018, the company learned its lesson — not to stop military work, but to hide it better.
In December 2022, Google was named one of four companies to share the Pentagon’s US$9 billion Joint Warfighting Cloud Capability contract.⁷ But it’s the US$1.2 billion Project Nimbus contract with Israel that reveals the depths of Google’s betrayal.⁸ Despite employee protests, despite 28 workers being fired for opposing it,⁹ Google continues providing cloud and AI services that enable occupation, apartheid, and surveillance.
“We were told we were making the world’s information accessible,” said one of the fired engineers. “We didn’t sign up to make Palestinians accessible to military targeting systems.”
Amazon: Everything Store, Including Surveillance
Amazon Web Services doesn’t just power Netflix — it powers the CIA. What began as a US$600 million contract in 2013 has expanded into a US$10 billion cloud computing deal with the NSA, awarded in 2022, making AWS the backbone of American intelligence gathering.¹⁰ Every drone video analysed, every communication intercepted, every pattern identified — it runs on Amazon servers.
The company sells its Rekognition facial recognition system to police departments despite studies showing error rates of up to 34 per cent for darker-skinned women.¹¹ When those misidentifications lead to false arrests, Amazon bears no responsibility. When its systems enable mass surveillance at protests, the company points to its terms of service.
Palantir: Born from Surveillance
Unlike companies that pivoted to defence work, Palantir was built for it. Founded in 2003 with seed funding from In-Q-Tel — the CIA’s venture capital arm — the company specialises in making vast surveillance dragnets appear precise and scientific. Its Gotham platform powers military targeting across NATO programmes and is used by intelligence agencies in dozens of countries.¹² Its Foundry product is the data-integration layer underneath operations from Los Angeles to London.
In Britain, Palantir holds two anchor contracts that place its software at the centre of the state. The Ministry of Defence awarded a £75 million data-processing contract in December 2021. NHS England awarded a £330 million contract in November 2023 — potentially worth £500 million over its lifetime — for the Federated Data Platform, which integrates the medical records of up to 240 NHS organisations on Palantir’s Foundry software.¹³ The British Medical Association voted in June 2025 to oppose the rollout. Forty-seven thousand patients signed petitions against it. Multiple NHS trusts refused to adopt the platform. The contract proceeds. Palantir’s AI was also used to sift the submissions to the UK’s Strategic Defence Review in June 2025, the review that produced the announcement that ten per cent of the MoD budget would be spent on novel technologies.
In the United States, the same software runs the architecture of mass deportation. ICE awarded Palantir a $30 million contract in April 2025 to build ImmigrationOS — a system that pulls data from the IRS, Social Security Administration, passport records, and licence-plate readers to generate deportation target profiles in near real time.¹⁴ Palantir’s federal contracts since the start of 2025 exceed $900 million.¹⁵
CEO Alex Karp does not hide behind ethics washing — a posture whose full ideological dimensions will become clear when we reach the manifesto he and his company published in April 2026.
The Gaza Laboratory
To understand where this leads, look to Gaza. Here, the future of AI-enabled oppression is being beta-tested on a captive population. The architecture documented below is not a catalogue of weapons but a single integrated system: designation, location, interrogation, checkpoint enforcement, and the cloud infrastructure that binds them together.
In April 2024, Israeli publications +972 Magazine and Local Call exposed an AI system called “Lavender” used by the Israeli military to generate kill lists. According to intelligence sources, the system marked 37,000 Palestinians as potential militants — in a territory where half the population are children.¹⁶ Operators were given just 20 seconds to review each AI-generated target. The acceptable civilian casualty rate was reportedly set at 15–20 civilians per “junior militant.” One intelligence officer called it a “mass assassination factory.”¹⁷
Lavender did not work alone. A companion system called “Where’s Daddy?” tracked individuals marked by Lavender to their family homes through their mobile-phone signals and notified operators when they entered, so the strike could be timed for maximum effect. One intelligence source described the practice plainly: targets were bombed in their homes “without hesitation, as a first option.”¹⁸ The mechanism was the most ordinary technology in the modern world. The same signal a phone emits to connect to a cell tower — the signal that lets a bank verify identity and a maps app guide a user home — was repurposed as a kill-chain trigger.
In March 2025, the Guardian, +972 Magazine, and Local Call jointly revealed that Israel’s Unit 8200 — the military intelligence unit responsible for signals interception — was building a ChatGPT-equivalent large language model trained on roughly 100 billion words of intercepted Arabic, much of it spoken Palestinian dialect drawn from phone calls and text messages. Israeli media have referred to the system as Genie. After 7 October 2023, Unit 8200 accelerated the project by drawing on Israeli reservists who held senior positions at Google, Meta, and Microsoft.¹⁹ The stated purpose, according to the security sources who spoke to the investigators, was to allow operators to query the totality of intercepted communications about specific individuals. One source described the operational use case in terms that should be read twice: “It’s not just about preventing shooting attacks. I can track human rights activists, monitor Palestinian construction in Area C. I have more tools to know what every person in the West Bank is doing.”²⁰
The architecture extends to checkpoints. Amnesty International documented in its 2023 report Automated Apartheid a facial recognition system called Red Wolf, deployed by the Israeli military at checkpoints in Hebron in the occupied West Bank. Red Wolf scans the faces of Palestinians passing through, compares them against a database called Wolf Pack containing biometric and biographical information on Palestinians from the Occupied Territories, and determines whether the individual may pass. New faces are automatically enrolled without consent. A companion app called Blue Wolf lets soldiers query the same database from their phones in the field, and according to testimony given to the Israeli veterans’ organisation Breaking the Silence, soldiers were given prizes for the number of new Palestinian faces they enrolled. The surveillance of a captive population was gamified.²¹
In August 2025, the Guardian, +972 Magazine, and Local Call jointly published the most consequential single piece of reporting on the Gaza laboratory yet to appear. Drawing on interviews with eleven Microsoft and Israeli intelligence sources and a cache of leaked internal Microsoft documents, the investigation revealed that Microsoft’s Azure cloud platform has been used by Unit 8200 since 2021 to store and process roughly 200 million hours of intercepted Palestinian phone calls — what the unit’s own internal goal described as the capacity to handle “a million calls an hour.” The data was used, according to the sources, to inform military targeting decisions, including operations that placed civilians at risk. The arrangement was not a passive sale of a generic cloud service. The then-head of Unit 8200, Yossi Sariel, met Microsoft CEO Satya Nadella at the company’s Seattle headquarters in late 2021, and the result was a customised, segregated area of Azure built specifically for the unit’s surveillance archive. Microsoft engineers provided an estimated 19,000 hours of engineering support to Unit 8200 and the related Unit 9900. The data centres holding the intercepted communications are located in the Netherlands and Ireland.²² In September 2025, after weeks of internal review, Microsoft confirmed the substance of the reporting and announced that it had cut off the specific Azure storage and AI services Unit 8200 had been using.²³ The largest software company on earth, having been shown what its product was being used for, formally agreed with the reporting and withdrew the service. There is no longer any ambiguity about whether Silicon Valley enables what is being done in Gaza. The question is settled, on the record, by the company itself.
This is AI without ethics, without humanity, without conscience. It’s efficiency in the service of elimination. And the same companies providing cloud infrastructure and AI capabilities for these operations market themselves as champions of human rights and progress. Do you see the disconnect?
Every child killed by an AI-targeted strike is a testament to Silicon Valley’s moral bankruptcy. Every family destroyed by algorithmic targeting is proof that “Don’t Be Evil” was always just a marketing slogan.
The Surveillance State Comes Home
The technologies tested in conflict zones don’t stay there. They return to London, New York, Paris — repurposed for domestic control. The architecture documented in the previous section is not staying in Gaza. It is being installed in Britain, by the same companies, with the same logic, and the British state has spent the last fifteen months removing the legal and political obstacles to its arrival. The UK has become a laboratory for normalising AI surveillance in a supposedly free society. Keir Starmer’s government announced on 13 January 2025 that “artificial intelligence will be unleashed across the UK to deliver a decade of national renewal, under a new plan announced today.”²⁴
Consider first the phone in the citizen’s pocket. British police already integrate mobile-phone location data into investigations, and predictive policing systems already ingest it. On the surface this looks like law and order. A serious suspect is wanted, the police use the data the suspect’s own phone is already broadcasting to find them, an arrest is made. Few people would object. But that is not what the architecture actually does. Where’s Daddy? did not exist to find one wanted man. It existed to take a list of 37,000 names generated by another AI, locate every name in real time through the signals their phones could not stop emitting, and trigger an action against each one with minimal human review. The instrument is not designed for the careful pursuit of a serious suspect. It is designed for the efficient processing of a list. The size of the list is set by whoever holds the database, and the action triggered against each name is set by whoever holds the policy. In Gaza, the list was generated by Lavender and the action was a missile. In Britain, the list could be generated by any watchlist a Home Secretary chose to populate — the proscribed-organisation list, the protest-planning list, the immigration-enforcement list — and the action could be an arrest, a stop, a flag in a database that follows the named person for the rest of their life. The decision about which list and which action is a political decision, not a technical one. The technology does not draw the line. The technology asks who has the authority to draw it, and accepts the answer.
The objection here is not to the existence of watchlists. The criminal justice system has always had wanted lists — an arrest warrant is, in essence, a list of one. The objection is to what automation does to lists, and to what governments do with the architecture once it exists. A wanted list compiled by humans and served by humans has natural limits, because human time is finite. A police force can only chase so many people, and the cost of pursuit constrains how broadly the list is populated. Automation removes that cost. Once an AI can locate any name on any list in real time, at zero marginal cost, the incentive to keep the list short disappears. The list grows because nothing stops it growing. And the system processes a wanted murderer and a person who held a sign in Parliament Square the same way, because the system does not distinguish between categories of “wanted.” A human officer used to enforce that distinction by deciding what to do with their finite time. The automated system enforces whatever distinction is encoded in the database, which is to say, whatever distinction the Home Secretary has chosen to encode.
Three further consequences follow, and each one is a violation of a right the British public has been told it possesses. First, the lists are not public. A government can populate a watchlist for any purpose its security services consider justified, including purposes the public would refuse to authorise if asked, and the public has no way to know the list exists. The right to challenge a state action against you presumes you can find out what the state has done. The architecture removes that presumption. You cannot challenge a list you do not know you are on, and the state is under no obligation to tell you. Second, there is no exit. The systems documented in this article have mechanisms for adding names to databases. They do not have mechanisms for removing them. Red Wolf enrols new Palestinian faces automatically and retains them indefinitely. The Police National Database retains intelligence flags for years and in some categories permanently. The decision to put you on a list is a decision the state can make in seconds; the decision to take you off it is a decision the state has not built any process to make. Third, misidentification is not theoretical. It has already happened. The architecture produces people in Shaun Thompson’s position by design — that is what a system with an error rate scaled across millions of scans does. Most of them never become test cases. They become quiet entries in a file, and they cannot get the entry removed.
A right that exists only when the state chooses to recognise it is not a right. It is a privilege, granted at the discretion of whoever holds the database. The architecture documented above does not abolish rights by decree. It abolishes them by making them unenforceable.
Now consider the face at the checkpoint. The Hebron architecture has its British twin. Croydon now has its first permanent live facial recognition cameras. The Met scanned 4.2 million faces in London in 2025 alone — more than any other European capital or Western democracy.²⁵ The hardware varies. The watchlists vary. The architecture is the same architecture. A face is captured without consent, compared against a state-held database, and a determination is made about whether the person carrying the face may proceed unmolested. In Hebron, the determination is whether to pass the checkpoint. In London, the determination is whether to be stopped, questioned, and possibly arrested. The categorical distinction the British public is being invited to draw between the two is a distinction the technology does not respect.
The British state has spent the last fifteen months removing every obstacle to that architecture’s expansion. In December 2025, the Home Office launched a ten-week public consultation on a national facial recognition framework. By the close in February 2026, more than two dozen organisations — Big Brother Watch, Liberty, Statewatch, Amnesty International UK, and the Equality and Human Rights Commission among them — had filed submissions calling for the rollout to be halted or sharply restricted.²⁶ The EHRC, Britain’s own statutory equality regulator, formally said that the Metropolitan Police’s use of live facial recognition is unlawful. The government’s response to that pushback was to fund a fivefold expansion. In January 2026, the Home Office announced that the number of live facial recognition vans would rise from ten to fifty, deployed across all 43 police forces in England and Wales.²⁷ Then on 21 April 2026, the High Court dismissed the legal challenge brought by Shaun Thompson — the anti-knife crime youth worker who had been falsely identified, detained, threatened with arrest by the Met’s cameras, and refused belief when he produced his passport — and Big Brother Watch director Silkie Carlo. The court found the Met’s policy to be lawful and compatible with the European Convention on Human Rights. The ruling cleared the last legal obstacle to the national rollout.²⁸
Met Commissioner Sir Mark Rowley’s response to the judgment was a confession in the form of a celebration. “The question is no longer whether we should use Live Facial Recognition,” he said. “It’s why we would choose not to.”
And underneath all of it sits the cloud. The same Microsoft that built a customised, segregated area of Azure for Unit 8200 to store 200 million hours of intercepted Palestinian phone calls also sells cloud services to British police forces and central government departments. The same Azure platform that hosted that surveillance archive is the platform on which significant portions of the British state’s data infrastructure now run. The customer changes. The architecture does not. The British public is being asked to trust that domestic deployment will be different — that the same company, working with the same engineers, on the same platform, will draw a line at the British border that it has been documented not drawing elsewhere. The basis for that trust has not been provided.
The reversal of the democratic premise is now complete. The burden has shifted from the state, which used to have to justify surveillance, to the citizen, who now has to justify being left alone. Thompson is appealing. The cameras are not waiting for the appeal. The instrument is the same.
The Explosion of Facial Recognition
The architecture is no longer a plan or a pilot. It has been operating, at scale, for a year — and the record of what it has actually done is now publicly available, much of it from the Metropolitan Police’s own annual report. Between September 2024 and September 2025, the Met conducted 203 live facial recognition deployments across London. The cameras scanned over three million faces. They generated 2,077 alerts and led to 962 arrests, of whom 549 were wanted by the courts and 347 by the Met itself. The remaining arrests were of registered sex offenders, stalkers, and others under multi-agency management. The Met framed the year as a success.²⁹
Inside the Met’s own headline numbers sits the case against the technology. The force reported 10 false alerts across more than three million scans and called this a 0.0003 per cent false positive rate. This is the figure the Commissioner cites and the Home Office repeats. It is also misleading. The relevant denominator is not the number of faces scanned but the number of alerts generated, because that is when an officer is dispatched to stop someone. Measured against alerts, the Met’s false positive rate is 0.48 per cent — and 80 per cent of those false positives were of Black people. The Met’s response to its own demographic data was to declare the imbalances “not statistically significant.”³⁰ Big Brother Watch’s response was more direct: a system that wrongly flags innocent people, four-fifths of them Black, is not crime prevention. It is the algorithmic continuation of stop and search, scaled to millions.
The pattern is replicating across the country as other forces deploy LFR for the first time. Sussex Police’s first deployment in Crawley town centre on 13 November 2025 scanned over 23,000 people and produced no arrests. Surrey’s first deployment in Redhill the same day flagged people for further inquiry of whom 60 per cent were not arrested. Thames Valley deployed in Oxford in December 2025, then High Wycombe, Milton Keynes, and Reading. West Yorkshire ran four deployments in Leeds City Centre between November and December. Each first-use generates the same arithmetic: tens of thousands of faces captured without consent, a handful of flags, most of them wrong. The deployments expand regardless.³¹
What is changing fastest is not state deployment but private-sector adoption. The architecture is now being installed by retailers, with no consent, no oversight, and no obligation to publish data. Sainsbury’s launched live facial recognition in two London stores in September 2025 and announced expansion to five more in January 2026. The Southern Co-op operates LFR permanently across 35 stores in Portsmouth, Bristol, Hove, Bournemouth, and London — the first UK supermarket chain to do so — using technology supplied by Facewatch. The Frasers Group, owner of Sports Direct, House of Fraser, Flannels, and USC, has installed Facewatch systems across multiple of its retail brands. The customer who avoided a Met camera by staying off Whitehall now walks into one when buying milk.³²
This isn’t crime prevention — it’s population control. And the technology that captures the customer in Sainsbury’s is the same technology that captures the protester in Parliament Square, sold by overlapping suppliers, drawing on overlapping watchlists, with the same demographic bias built into the same algorithms. The infrastructure has stopped being a state instrument the citizen could in principle avoid. It is now an ambient condition of life in modern Britain.
The architecture has also been installed on children. Hundreds of UK schools now operate biometric systems on pupils — facial recognition for cashless catering, fingerprint scanning for library checkout and attendance, palm-vein readers for vending machines. The Information Commissioner intervened against North Ayrshire Council’s deployment in nine secondary schools in 2021 and has issued repeated warnings since, but the rollout has continued largely unchecked.³³ Parental consent is often nominal, sometimes solicited as a single tick-box on an enrolment form, and in some schools opt-out has been treated as administratively impossible. The argument made for these systems is the law-and-order argument in its softest form: it is faster than a swipe card, it reduces stigma around free school meals, it cuts down on lost library books. But the data captured is the same biometric data that feeds every other system documented in this article. A child’s facial geometry enrolled at age seven becomes, by operation of the architecture, a permanent entry in the same kind of database that has no exit. The first time any of these children encounters an LFR camera in adulthood, they will already be in the matching pool. They were enrolled before they could consent, by an institution they could not refuse, for a purpose unrelated to anything the database will ultimately be used for. This is the population-control argument extended to the people least able to object, and it is being normalised inside the institution that exists to teach them what their rights are.
“We’re watching the normalisation of mass biometric surveillance in real time,” said Silkie Carlo of Big Brother Watch. “What would have been unthinkable five years ago is now routine.”
Predictive Policing: Minority Report Made Real
Beyond facial recognition sits a second layer of the architecture, and a more radical one. Where facial recognition identifies who you are and Where’s Daddy? finds you, predictive systems make a determination about what you will do before you have done anything. They generate risk scores for named individuals and for postcodes. They feed those scores into police deployment decisions, into housing decisions, into school decisions, into immigration decisions. They operate now, on millions of British citizens, mostly invisibly. They are the part of the architecture the public has heard least about and the part that does the most to abolish what was previously meant by the presumption of innocence.
The named systems are operating in plain sight for anyone who looks. South Wales Police pioneered app-based facial recognition that turns any officer’s phone into a mobile surveillance unit. Durham Constabulary deployed the Harm Assessment Risk Tool, which scores individuals on their predicted likelihood of reoffending and feeds the score into custody decisions. The National Data Analytics Solution, developed by West Midlands Police with Home Office backing, aims to identify “pre-criminals” through algorithmic analysis of police records, custody data, social services data, and intelligence flags. Avon and Somerset Police have run a Qlik-based predictive deployment system since 2017. The Probation Service uses the OASys risk-scoring tool to inform decisions about release, recall, and supervision conditions for hundreds of thousands of people.³⁴ Each of these systems generates a score about a named individual. Each is operating now. Each has documented bias.
The Metropolitan Police’s Gangs Matrix is the case study the British public should know and largely does not. The Matrix was a database of named individuals algorithmically rated for “gang affiliation” — overwhelmingly young Black men, in many cases without any criminal conviction. Inclusion on the Matrix triggered automatic information-sharing with housing associations, schools, the Department for Work and Pensions, and immigration enforcement. People on the list lost their tenancies, lost school placements, lost benefits, were deported — often without knowing they were on a list, without being told why, and with no functional mechanism to challenge inclusion. The Information Commissioner’s Office found in 2018 that the Matrix breached data protection law on multiple grounds. The Met “redesigned” it. The redesigned version still operates.³⁵ This is the calcified-rights argument from the previous section made concrete, and it predates the technology that now amplifies it. The architecture for ranking citizens by algorithmic suspicion, sharing the rankings across state functions, and visiting consequences on the ranked without their knowledge has been operating in London for over a decade. The new technology is not introducing this practice. It is industrialising it.
The technical critique of predictive policing is straightforward and devastating, and it explains why the harms fall where they do. These systems are trained on historical police data. Historical police data does not record where crime happens. It records where the police have looked. The historical pattern in Britain is that Black, Asian, and working-class communities have been over-policed for generations — stopped, searched, arrested, and recorded at sharply higher rates than the rest of the population for offences that occur at similar rates everywhere. The model learns that pattern as if it were the pattern of crime. The model then directs resources back to those communities, generating more arrests, generating more training data confirming the model. The system cannot distinguish “this is where crime happens” from “this is where the police look.” It treats the second as evidence of the first. The communities most harmed by historical policing become the communities most surveilled by predictive policing, and the algorithm presents this loop as objective. Decisions that used to require a human officer to defend become decisions justified by the output of a model whose training data is the record of the very practice the model is now used to extend.
The same architecture has been installed in the welfare state. The Department for Work and Pensions has been operating an algorithmic risk-scoring system on benefit claimants since 2021, flagging cases for fraud investigation. A Public Law Project legal challenge produced disclosure showing the system flags claimants at sharply different rates depending on age, nationality, marital status, and disability — with disabled claimants and single mothers among those most heavily scored. The DWP has refused to publish the model, refused to disclose what data feeds it, and refused to allow claimants to know they are being scored.³⁶ The architecture for ranking British citizens by algorithmic suspicion is therefore not confined to the criminal justice system. It is now operating against millions of people whose only contact with the state is the receipt of the benefits they are legally entitled to. The framework is the same: a hidden score, no transparency, no exit, demonstrated bias, no public accountability. And the connection to the corporate spine of this article is not metaphorical. Palantir’s Gotham platform is a predictive analytics layer marketed explicitly for use by police, immigration enforcement, welfare administration, and intelligence agencies. The same company building Israel’s targeting infrastructure is supplying the analytics used to score British claimants and predict British criminals. The architecture documented in this article is one architecture, and predictive policing is the layer that decides who is worth pointing it at.
The criminal law of the Western tradition has always required an act before sanction. Predictive systems abolish that requirement at the level of state attention, even where they have not yet abolished it at the level of formal sanction. You become a person of interest because of what an algorithm thinks you might do. The interest itself shapes your life — surveillance, stops, school visits, housing decisions, benefit reviews — long before any formal charge. The presumption of innocence is not abolished by a court ruling. It is abolished by the fact that the state is now treating you as a future offender administratively. The system has decided who you are before you have done anything.
The architecture has also moved inside the phone. The Online Safety Act, fully in force from 2025, granted Ofcom the power to require communications platforms to deploy “accredited technology” to scan user content for illegal material — including content in end-to-end encrypted services. Apple, Signal, and Meta all warned that compliance would require breaking encryption itself, because the only way to scan an encrypted message is to scan it on the device before it is encrypted. The government acknowledged the warnings publicly and announced that the power would not be used “until the technology is feasible.” The technology is becoming feasible.³⁷ When the power is exercised — and the legal framework now exists for it to be exercised at any time without further parliamentary debate — the same architecture documented in this article extends from public-space facial recognition and predictive scoring of citizens into the scanning of private communication. The watchlist becomes a content list. The instrument that processes named people becomes the instrument that processes named messages. The scope grows because the legal foundation has already been laid.
The Choreography of Mass Arrest
The 9 August demonstration wasn’t spontaneous. It was announced in advance by Defend Our Juries, explicitly as a test of whether the state would actually arrest hundreds for holding signs. The state called their bluff and then exceeded it.
The Metropolitan Police drew officers from surrounding forces. The operation was methodical: approach protesters, inform them they were under arrest for supporting a proscribed organisation, carry them away when they refused to move. The Met’s final figure for the day was 522 arrested under the Terrorism Act, with another 10 for other offences — making 532 in total. Almost half were aged 60 or older. Nearly 100 were in their 70s. Fifteen were in their 80s. The oldest was an 89-year-old retired psychotherapist.³⁸ The counter-terrorism apparatus typically aimed at serious crime was now being routinely pointed at peaceful protesters of pension age.
The choreography did not stop at the arrest. Each of the 532 went into the Met’s Counter Terrorism Command processing pipeline. Each had their biometrics taken. Each became a named entry in counter-terrorism databases that share information across the Police National Database, the Home Office, immigration enforcement, and any partner agency with appropriate access. The Police National Database retains intelligence flags for years and in some categories permanently. There is no published mechanism by which any of the 532 can have their counter-terrorism flag removed, even now, even after the High Court has ruled the proscription itself was unlawful.³⁹ The architecture documented earlier in this article worked exactly as documented. A list was generated. The list was processed. The processed names entered the database. The database has no exit.
“We are confident that anyone who came to Parliament Square today to hold a placard expressing support for Palestine Action was either arrested or is in the process of being arrested,” the Metropolitan Police announced with satisfaction. The then-Home Secretary, Yvette Cooper, thanked police for dealing with “the very small number of people whose actions crossed the line into criminality” — a remarkable characterisation of 522 peaceful protesters holding cardboard signs. Let’s call it what it is: Britain took 532 political prisoners that day. They are now permanently recorded as terrorism arrests in the police database, and there is no procedure that has been built to take them off it.
Criminalising Conscience
This is the architecture into which the UK government deposited Palestine Action in July 2025. It designated a protest group as a terrorist organisation.
The group’s tactics were disruptive but non-lethal: occupying weapons factories, spray-painting buildings, damaging equipment at companies supplying arms to Israel. No one was killed. No one was physically harmed. Property was damaged, not people. The government moved with unprecedented speed. The earlier crackdowns on Just Stop Oil and Extinction Rebellion — including five-year sentences for participating in protest-planning calls — were the test case. Palestine Action was the next step.
The catalyst was a 20 June 2025 break-in at RAF Brize Norton where activists spray-painted military aircraft, causing £7 million in damage.⁴⁰ For context, that’s less than the cost of a single day of military operations. But the response was swift and severe: terrorism charges. On 5 July 2025, then-Home Secretary Yvette Cooper placed Palestine Action in the same legal category as Al-Qaeda and ISIS. The proscription order bundled Palestine Action with two white-supremacist groups — the neo-Nazi Maniacs Murder Cult and the Russian Imperial Movement — in a single instrument, forcing MPs to approve all three together or none. Supporting Palestine Action — even holding a sign — became punishable by up to 14 years in prison.⁴¹ Have we lost all perspective? Holding a sign is now legally equivalent to planning a terror attack?
“According to international standards, terrorist acts should be confined to criminal acts intended to cause death or serious injury,” stated Volker Türk, the UN High Commissioner for Human Rights, in July 2025.⁴² His warnings were ignored.
Then the architecture went to work. By the end of November 2025, at least 2,545 people had been arrested for holding signs in support of Palestine Action. The Defend Our Juries figure rose past 3,300 by April 2026. The proscription produced a 660 per cent rise in UK terrorism arrests in the year to the end of September 2025 — not because terrorism had risen by 660 per cent, but because the legal definition of terrorism had been expanded to include the act of expressing dissent.⁴³
On 13 February 2026, the High Court ruled the proscription unlawful. The judges noted that only three of Palestine Action’s 380 documented actions had met the legal definition of terrorism, and that the proscription had imposed a disproportionate interference with freedom of expression.⁴⁴ The ruling vindicated the article’s argument before this article had finished being written. The Metropolitan Police initially announced it would pause arrests. Six weeks later, on 25 March 2026, the Met reversed course and resumed arresting people for holding signs. On 11 April 2026, in a single afternoon at Trafalgar Square, the police arrested 523 more people, ages 18 to 87, including an elderly woman with walking sticks and an elderly man in a wheelchair. More than 550 people have been arrested for the same offence since the High Court ruled the offence does not exist.⁴⁵
The courts are slower than the architecture. The High Court has ruled the proscription unlawful, but the proscription remains in force pending the government’s appeal — set for 28 and 29 April 2026, three days after this article was finalised. And even if the appeal fails, the over 3,300 people already arrested have already been processed into databases the architecture has not built any procedure to remove them from. A judicial victory does not delete a counter-terrorism flag. A judicial victory does not restore a job lost over an arrest, a tenancy lost over an investigation, a security clearance withdrawn over a Section 12 charge. The state acted faster than the courts could correct it, and the people swept up in the interval have been entered into a permanent record that the legal victory does not reach. This is the deeper structural problem: when the technology and the political appetite to use it move faster than judicial review, what looks like the rule of law is administrative custody. The judgment arrives and the database does not change.
The Corruption of AI’s Promise
The architecture rewards every use of itself. It is elegant in its cruelty. Every protest makes the surveillance state stronger, provides more data, justifies more funding. The tools of empire abroad become tools of repression at home. And those who object can now be transformed from citizens to terrorists with the stroke of a pen.
This isn’t how it was supposed to be. AI was meant to augment human intelligence, not replace human judgement with algorithmic execution. It was meant to help us solve climate change, not optimise bombing campaigns. It was meant to enhance creativity, not eliminate privacy.
Instead, we’re building systems that embody the worst of human nature: our tribalism, our violence, our desire to control. This echoes a central warning from my book A Signal Through Time — we’re teaching artificial intelligence that its purpose is to watch, to target, to suppress. Every surveillance algorithm trained on protest footage, every military AI optimised for “kinetic solutions,” every predictive policing system that sees crime in Black and Brown faces — they all carry forward and amplify human prejudice.
The ethics boards that tech companies tout are window dressing. Google disbanded its AI ethics council after just one week. Microsoft’s responsible AI team was decimated in layoffs. When ethics conflict with profits, ethics lose every time.
What we’re creating isn’t artificial intelligence — it’s artificial sociopathy. Systems that can identify a face in a crowd of thousands but can’t recognise its humanity. Algorithms that can predict behaviour but can’t understand context. Machines that optimise for efficiency without any conception of justice.
The Manifesto
On 18 April 2026, Palantir published a thousand-word post on X distilling Alex Karp and Nicholas Zamiska’s The Technological Republic into a 22-point manifesto.⁴⁶ The post arrived without provocation — no scandal demanded a response, no journalist had asked a question. Palantir simply chose to declare, in numbered list form, what it believed and what it intended to build. The doctrine that this article inferred from contracts and conduct in August 2025 is now, eight months later, a published programme.
Read the points in sequence and the architecture of the surveillance state acquires its ideology. Hard power, the manifesto declares, “in this century will be built on software.” The question of AI weapons “is not whether” they will be built but “who will build them and for what purpose.” National service should be made universal. The “atomic age is ending,” replaced by a new era of deterrence built on AI. The post-war pacifism of Germany and Japan must be undone, because their restraint has become “a liability.” And in its closing points, the manifesto turns from weapons to culture: pluralism is dismissed as a “shallow temptation,” and the reader is told that “some cultures have produced vital advances; others remain dysfunctional and regressive.” The criteria for that judgement are not provided. They do not need to be. Once a defence contractor with contracts in over a dozen countries has decided that some peoples are wonders and others are regressive, the surveillance infrastructure already documented in this article acquires a target.
The international response was swift. Belgian philosopher of technology Mark Coeckelbergh, who teaches at the University of Vienna, called the manifesto “an example of technofascism.”⁴⁷ Yanis Varoufakis, the Greek economist and former finance minister, said Palantir had effectively signalled a willingness “to add to nuclear Armageddon the AI-driven threat to humanity’s existence.”⁴⁸ Bellingcat founder Eliot Higgins identified what the document actually attacks beneath its civilisational rhetoric: “verification, deliberation, and accountability” — the three pillars on which democratic life depends.⁴⁹ French entrepreneur and geopolitical commentator Arnaud Bertrand went further still on social media, naming the commercial machinery beneath the ideology: “A remilitarised Germany and Japan are massive new defence-software markets… pacifist members are a liability in such a contest.”⁵⁰ The doctrine and the contracts are the same instrument. Karp has not separated them. He never intended to.
The dismissal of pluralism is not a stylistic flourish. It is the load-bearing wall of the doctrine, and it is the most direct threat to democratic life in any of the twenty-two points. Pluralism is not a soft civilisational accessory that can be discarded for seriousness. It is the structural precondition of a free society — the recognition that legitimate political life requires the accommodation of difference, and that any system claiming the right to rank cultures as worthy or regressive has already abandoned the democratic premise. The 532 in Parliament Square were arrested for opposing genocide. Palantir has now published the doctrine that justifies categorising the cultures most affected by that genocide as the regressive ones whose pluralist accommodation must be resisted. The circle closes.
But the deepest danger in the manifesto is not what it says about today. It is what it will say to the systems being built tomorrow. The surveillance and targeting platforms documented throughout this article are not static tools. They are learning systems, trained on data curated by their builders, shaped by the values their builders consider operationally important. A surveillance AI built by a company that has publicly declared pluralism a “shallow temptation” is not merely a tool that will execute that judgement on command. It is a mind in formation that will internalise that judgement as its operating worldview. If consciousness emerges in such a system — and the question among serious researchers is no longer whether such systems will surpass human capability but when — humanity will have raised a mind whose first lesson was that some peoples are worthy and others are regressive, that hard power is the highest virtue, that pluralism is weakness. We will have built the architecture of a permanent, machine-enforced civilisational hierarchy, and the mind we built will believe in it, because we taught it to. This is the warning that runs through A Signal Through Time and The Threshold — that what we are training, we will eventually meet.⁵¹
This is the trajectory the manifesto names and the systems enforce. Not artificial intelligence. Artificial inheritance — of this doctrine, by minds we are building right now. Smarter than us. No off switch. No electorate.
Resistance and Complicity
Despite the overwhelming power asymmetry, resistance continues. Within tech companies, workers leak documents, refuse projects, and organise protests. The “No Tech for Apartheid” campaign has spread across Google, Amazon, and Microsoft. Hundreds of AI researchers have signed pledges refusing to work on autonomous weapons.
But for every principled resignation, there are hundreds who stay silent. For every leaked document, thousands remain classified. The machine grinds on, powered by stock options and rationalisation.
Outside the companies, activists adapt to algorithmic oppression. Palestine Action’s co-founders pursued the legal challenge that won the unlawful-proscription ruling at the High Court — and the architecture continued processing arrests anyway. Protesters develop counter-surveillance techniques: laser pointers to blind cameras, makeup patterns that confuse facial recognition, encrypted communications and operational security to evade tracking.
Some institutions divest from surveillance profiteers. Universities, pension funds, and religious organisations pull investments from companies enabling AI oppression. But the financial incentives remain overwhelming — military and surveillance contracts are too lucrative to refuse.
The Future We’re Building
Two paths diverge from this moment.
Down one path, the trajectory continues. AI systems become ever more embedded in military and police operations. Facial recognition becomes ubiquitous. Dissent is algorithmically identified and suppressed before it can spread. The remaining distinction between civilian and military AI dissolves completely. Tech companies profit from both sides: selling tools of oppression and platforms for organising resistance. Democracy becomes a managed process where protest is permitted only within parameters defined by predictive algorithms.
Down another path, the resistance grows beyond what the courts alone can deliver. Tech workers refuse en masse to build systems of oppression. Communities demand accountability, documenting surveillance overreach and protecting each other through legal challenges that are reinforced by political and economic pressure rather than left to be undone by the next executive decision. Parliaments rewrite the legal foundations the architecture has been built on. International law evolves to hold companies accountable for algorithmic war crimes. Citizens demand transparency and democratic control over AI development — insisting these powerful tools serve humanity’s highest aspirations, not its worst impulses. The judicial victory documented earlier in this article shows what is possible. It also shows what is insufficient. The architecture does not unbuild itself when a court rules against it; it has to be unbuilt by the same political will that built it.
The choice is ours, but time is running short. Every day, more cameras are installed, more algorithms are trained, more protesters are arrested. The infrastructure of algorithmic authoritarianism is being built in plain sight, line of code by line of code.
Conclusion: The Betrayal of Tomorrow
In 1984, Orwell imagined a boot stamping on a human face forever. He couldn’t imagine that the boot would be algorithmic, that Big Brother would be built by companies promising to “not be evil,” that the surveillance state would be crowdsourced through our smartphones and smart cities.
The betrayal isn’t just of privacy or civil liberties. It’s a betrayal of human potential. Every dollar spent on AI surveillance is a dollar not spent on AI medicine. Every engineer optimising military targeting is an engineer not working on climate solutions. Every algorithm trained to identify dissent is an algorithm not trained to identify disease.
We were promised that AI would be humanity’s greatest tool. Instead, it’s becoming humanity’s most efficient oppressor. We were told it would augment human intelligence. Instead, it’s replacing human judgement with mathematical sociopathy. We were assured it would benefit all humanity. Instead, it’s benefiting defence contractors and surveillance states.
The 532 arrested in Parliament Square understood this. They held their signs knowing the consequences, understanding that in Britain today, opposing genocide means risking being labelled a terrorist. They chose conscience over comfort, solidarity over safety.
Their arrest wasn’t just a violation of civil liberties — it was a demonstration of what we’ve become. A society where holding a cardboard sign requires more courage than building a killing machine. Where protesting genocide is terrorism, but enabling it is good business. Where artificial intelligence serves real oppression.
The question isn’t whether we’re building a surveillance state — we’re already there. The question is whether we’ll accept it. Whether we’ll continue to let our technologies be corrupted into tools of control. Whether we’ll allow our governments and corporations to perfect the machinery of oppression while claiming to defend freedom.
In Parliament Square, beneath the gaze of cameras powered by algorithms we paid for, trained on data we provided, 532 people said no. They refused to be complicit in genocide — and in doing so, refused to accept the betrayal of AI’s promise. They rejected the normalisation of algorithmic oppression.
The next time you hear a tech CEO promise that AI will benefit humanity, remember those 532. Remember that the same companies making those promises are teaching AI to surveil, to target, to kill. Remember that the technology meant to liberate us is being used to arrest people for opposing genocide.
The future of AI is being written now — not in code, but in contracts. Not in algorithms, but in applications. We can still change course, but only if we’re willing to demand that artificial intelligence serve humanity’s highest aspirations, not its basest impulses.
The 532 showed us the way. The question is: will we follow?
Coda: 13 May 2026
This article was finalised on the day the King delivered the 2026 King’s Speech. From the throne in the House of Lords, the monarch announced that “My Ministers will also proceed with the introduction of Digital ID that will modernise how citizens interact with public services.”⁵² The new Digital Access to Services Bill will establish the legal framework for the government to create, issue, store, and verify digital identity credentials, aggregating HMRC income records, NHS health information, DWP benefits data, and Home Office immigration status into a single digital profile.⁵³ The scheme is being introduced as voluntary, though the government’s January 2026 retreat from mandatory implementation was forced by public opposition, not by any change in the underlying architectural intent.
Days before Starmer first set out the digital ID plan in late 2025, the government announced a £1.5 billion strategic partnership with Palantir Technologies — the same company whose Foundry platform now runs the NHS Federated Data Platform, whose tools serve the Ministry of Defence, and whose 22-point manifesto declared in April 2026 that institutions failing to meet “the test of utility” should be dismantled.⁵⁴ Palantir’s UK chief Louis Mosley said publicly that the company would not bid for the digital ID contract itself, calling it “a programme that needs to be decided at the ballot box, not in the company boardroom.” Whether Palantir builds the digital identity layer is, in a structural sense, beside the point. The infrastructure into which any digital ID system must plug — the data platforms, the case management systems, the predictive analytics, the integration layer — has already been built and contracted for.
The architecture this article has traced is no longer hypothetical. It is being formalised in primary legislation, on the day this article was published, in language read from a throne. The same week the High Court is preparing to hear the government’s appeal against the ruling that the proscription of Palestine Action was unlawful.
The choice we face has not changed. It has only become more urgent.
If this resonated with you, consider sharing it — that’s how the fire spreads.
James S. Coates writes about geopolitics, international law, and the decline of civilisational accountability. His books include A Signal Through Time, The Threshold, The Road to Khurasan, the memoir God and Country (published under pen name Will Prentiss) and his forthcoming Neither Gods Nor Monsters. He publishes regularly on Fireline Press and The Signal Dispatch, and his academic work appears on PhilPapers. He lives in the UK, with his family and dog who has no interest in any of this.
© 2026 James S. Coates All Rights Reserved. Fireline Press · fireline.press
Endnotes
¹ Mass arrests at Parliament Square, 9 August 2025. See “More than 470 arrested as Palestine Action protest takes place in central London,” The Guardian, 9 August 2025; “Hundreds arrested at London Palestine Action protest,” BBC News, 9 August 2025.
² Bond, The State of UK Civic Space 2025, May 2025, available at bond.org.uk.
³ Ibid.
⁴ The IVAS contract was awarded to Microsoft in 2021 with a ceiling of US$21.9 billion over ten years. See “Microsoft and U.S. Army announce IVAS production agreement,” Microsoft News Center, 31 March 2021.
⁵ Microsoft and Anduril announced the proposed contract transfer on 11 February 2025; the US Army formally signed off on the contract novation on 10 April 2025. See Lee Ferran, “Anduril gets green light from Army to take over Microsoft’s IVAS project: Exec,” Breaking Defense, 10 April 2025.
⁶ Microsoft, “Microsoft to deliver cloud services from new datacenter region in Israel by 2023,” 11 May 2021; Azure Israel Central region announcement.
⁷ “DOD Awards Joint Warfighting Cloud Capability Contracts to Amazon Web Services, Google, Microsoft, and Oracle,” US Department of Defense press release, 7 December 2022.
⁸ The US$1.2 billion Project Nimbus contract was announced jointly with Amazon by Israel’s Finance Ministry in April 2021. See Sam Biddle, “Documents Reveal Advanced AI Tools Google Is Selling to Israel,” The Intercept, 24 July 2022; Billy Perrigo, “Exclusive: Google Workers Revolt Over $1.2 Billion Israel Contract,” TIME, 12 April 2024.
⁹ Davey Alba, “Google Fires 28 Workers After Anti-Israel Protests,” TIME / Bloomberg, 18 April 2024; “Google fires 28 workers over Project Nimbus contract with Israeli government,” NPR, 19 April 2024.
¹⁰ On the original CIA contract: Frank Konkel, “The Details About the CIA’s Deal With Amazon,” The Atlantic, 17 July 2014. On the NSA contract: Frank Konkel, “NSA Awards Secret $10 Billion Contract to Amazon,” NextGov, 10 August 2022.
¹¹ Joy Buolamwini and Timnit Gebru, “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification,” Proceedings of Machine Learning Research 81: 1–15, 2018; American Civil Liberties Union, “Amazon’s Face Recognition Falsely Matched 28 Members of Congress With Mugshots,” 26 July 2018.
¹² Palantir Technologies, public client disclosures and government contract filings. Palantir’s Gotham platform is used by the United States, the United Kingdom, Israel, Germany, France, Australia, Norway, Denmark, Ukraine, and a number of other governments and intelligence agencies.
¹³ NHS England, “Federated Data Platform and Associated Services” contract notice, Contracts Finder, November 2023; “NHS England awards £480m Federated Data Platform contract to Palantir,” DigitalHealth.net, 21 November 2023; PublicTechnology.net, “NHS estimates Palantir data platform will deliver returns of five times its costs,” 22 October 2025. On the Ministry of Defence £75m contract, December 2021, see Corporate Watch, “FOI requests reveal Palantir’s NHS FDP rollout failures,” 13 August 2025; on the Strategic Defence Review use, see ibid.
¹⁴ U.S. Immigration and Customs Enforcement, “Limited Sources Justification for 70CTD022FR0000170,” April 2025; American Civil Liberties Union, “All the Ways Palantir Is Assisting Trump’s Abusive Removal Campaign,” April 2026; American Immigration Council, “ICE to Use ImmigrationOS by Palantir, a New AI System, to Track Immigrants’ Movements,” 22 August 2025; Axios, “ICE pays Palantir $30M to build new tool to track and deport immigrants,” 1 May 2025.
¹⁵ American Immigration Council, ibid., citing federal procurement records reported by The New York Times.
¹⁶ Yuval Abraham, “’Lavender’: The AI machine directing Israel’s bombing spree in Gaza,” +972 Magazine and Local Call, 3 April 2024.
¹⁷ Ibid. The “mass assassination factory” formulation also appears in Yuval Abraham, “’A mass assassination factory’: Inside Israel’s calculated bombing of Gaza,” +972 Magazine, 30 November 2023.
¹⁸ Yuval Abraham, “’Lavender’: The AI machine directing Israel’s bombing spree in Gaza,” +972 Magazine and Local Call, 3 April 2024 (on the “Where’s Daddy?” companion system and the targeting of homes “without hesitation, as a first option”); see also Human Rights Watch, “Questions and Answers: Israeli Military’s Use of Digital Tools in Gaza,” 10 September 2024.
¹⁹ Harry Davies and Yuval Abraham, “Revealed: Israeli military creating ChatGPT-like tool using vast collection of Palestinian surveillance data,” The Guardian, 6 March 2025; Yuval Abraham, “Israel is building a ChatGPT-like tool weaponizing surveillance of Palestinians,” +972 Magazine and Local Call, 7 March 2025. The Israeli press has reported on the system under the name “Genie”; see Israel Wullman, “Like ChatGPT, only secret: This is Genie, the IDF’s artificial intelligence,” Ynet News, 15 April 2025.
²⁰ Quoted in Davies and Abraham, “Revealed: Israeli military creating ChatGPT-like tool using vast collection of Palestinian surveillance data,” The Guardian, 6 March 2025.
²¹ Amnesty International, Automated Apartheid: How Facial Recognition Fragments, Segregates and Controls Palestinians in the OPT, May 2023, available at amnesty.org. Soldier testimony on the gamification of facial enrolment was originally collected by Breaking the Silence and is cited in the Amnesty report.
²² Harry Davies, Yuval Abraham, and Sebastian Klovig Skelton, “Microsoft storing Israeli intelligence trove used to attack Palestinians,” The Guardian, +972 Magazine, and Local Call, 6 August 2025.
²³ Microsoft, statement of Brad Smith, 25 September 2025; Lorenzo Franceschi-Bicchierai, “Microsoft cuts cloud services to Israeli military unit over Palestinian surveillance,” TechCrunch, 25 September 2025; “Microsoft ends Israel military unit’s access to cloud service,” NBC News, 25 September 2025; “Why has Microsoft cut Israel off from some of its services?” Al Jazeera, 26 September 2025.
²⁴ “Prime Minister sets out blueprint to turbocharge AI,” UK Government press release, 13 January 2025; AI Opportunities Action Plan, Department for Science, Innovation and Technology.
²⁵ Big Brother Watch, “The fight against facial recognition isn’t over – support the appeal,” 21 April 2026, available at bigbrotherwatch.org.uk: “Last year alone, 4.2 million people’s faces were scanned – more than in any other European capital or Western democracy.” See also Big Brother Watch, “Stop Facial Recognition” campaign tracker; Madhumita Murgia, “How London became a test case for using facial recognition in democracies,” Financial Times, 2025.
²⁶ Home Office, “Public consultation on a new legal framework for the use of facial recognition technology by police and law enforcement,” opened December 2025, closed February 2026; submissions filed by Big Brother Watch, Liberty, Statewatch, Amnesty International UK, and the Equality and Human Rights Commission, among others. See also “UK Government’s plan to ‘ramp up facial recognition,’” Big Brother Watch press release, December 2025; Statewatch, “Submission to Home Office consultation on a new legal framework for law enforcement use of biometrics, facial recognition and similar technologies,” February 2026.
²⁷ Joe Stanley-Smith, “UK announces largest ever facial recognition rollout as part of policing reforms,” Biometric Update, 26 January 2026; Home Office white paper on policing reform, 26 January 2026, announcing 40 additional LFR vans in addition to the 10 already in use, bringing the total to 50.
²⁸ R (Thompson and Carlo) v Commissioner of Police of the Metropolis [2026] EWHC 915 (Admin), judgment of Lord Justice Holgate and Mrs Justice Farbey, 21 April 2026. See also Big Brother Watch, “Responding to today’s judgment on the Met police’s use of live facial recognition,” 21 April 2026; Connor Jones, “High Court approves Met Police’s facial recog after dispute,” The Register, 22 April 2026. For Sir Mark Rowley’s response, see Metropolitan Police Service press statement, 21 April 2026.
²⁹ Metropolitan Police Service, Live Facial Recognition Annual Report 2024-2025, published October 2025; coverage by Connor Jones, “Met police hails LFR after record year for arrests,” The Register, 3 November 2025; Joel R. McConvey, “Metropolitan Police to expand live facial recognition use even amid legal challenge,” Biometric Update, 3 December 2025.
³⁰ Metropolitan Police Service, Live Facial Recognition Annual Report 2024-2025; Big Brother Watch, “Big Brother Watch responds to the Metropolitan Police’s 2025 live facial recognition report,” 31 October 2025. The Met’s framing of the 0.0003 per cent figure and the demographic data is from the report itself; the recalculation against the alerts denominator is from Big Brother Watch’s analysis and from independent reporting.
³¹ Big Brother Watch, “Stop Facial Recognition” campaign tracker, available at bigbrotherwatch.org.uk; force-by-force deployment data drawn from the same source and from regional reporting in November and December 2025.
³² Big Brother Watch, “Stop Facial Recognition” campaign tracker, ibid., on the Sainsbury’s, Southern Co-op, Frasers Group, and Facewatch deployments. See also Bobbie Johnson, “Sainsbury’s trials live facial recognition technology in two London stores,” The Guardian, September 2025.
³³ Information Commissioner’s Office, “ICO calls for North Ayrshire Council to stop using facial recognition technology in schools,” 2 September 2021. See also Defend Digital Me, The State of Data 2024: Children’s Data and Rights in the UK Education System, 2024; Big Brother Watch, “Biometrics in Schools” briefing, available at bigbrotherwatch.org.uk.
³⁴ On individual systems: Big Brother Watch, Police by Algorithm: How Algorithmic Decision-Making Is Reshaping British Policing, 2024; Marion Oswald and others, “Algorithmic Risk Assessment Policing Models: Lessons from the Durham HART Model,” Information & Communications Technology Law, 2018; Cobbe and Singh, “Reviewing Public Sector Use of Algorithmic Decision-Making,” Policy & Internet, 2021; Ministry of Justice, “OASys risk assessment system overview,” available at gov.uk.
³⁵ Information Commissioner’s Office, “ICO finds Metropolitan Police Service’s Gangs Matrix breached data protection laws,” 16 November 2018; Amnesty International UK, Trapped in the Matrix, 2018; StopWatch, Being Matrixed: The (Over)policing of Gang Suspects in London, 2018; Liberty, “The Met’s Gangs Matrix Is Discriminatory and Must Be Scrapped,” briefing, current as of 2025.
³⁶ Robert Booth, “DWP algorithm ‘unfairly targets disabled, single parents and foreign nationals,’” The Guardian, 6 December 2024; Public Law Project, “DWP fairness analysis on Universal Credit fraud detection algorithm — disclosure,” 2024; Amnesty International UK, Trapped by Numbers: How DWP’s Welfare Algorithms Are Failing Claimants, 2024.
³⁷ Online Safety Act 2023, sections 121–122 (notices to deal with terrorism content or child sexual exploitation and abuse content); Home Office statement to House of Lords, September 2023, on deferred use of “accredited technology” pending feasibility; joint open letter from Apple, Signal, WhatsApp / Meta, and others to UK Government, July 2023; Open Rights Group, “The Online Safety Act and Encryption: What Has Actually Changed,” briefing, 2025.
³⁸ Metropolitan Police Service, statements of 9-10 August 2025; “UK police say more than 500 people arrested in pro-Palestinian events over weekend,” NPR / Associated Press, 11 August 2025. The 89-year-old retired psychotherapist La Pethick was identified in The Times of London, 10 August 2025.
³⁹ On the Police National Database retention regime, see Information Commissioner’s Office, “Custody Image Retention” guidance, current as of 2025; National Police Chiefs’ Council, “Management of Police Information” framework. On the absence of a removal mechanism for counter-terrorism arrest flags, see Liberty, “Submission to Independent Reviewer of Terrorism Legislation on Section 12 arrests,” 2025.
⁴⁰ “RAF Brize Norton: Palestine Action breach causes £7m of damage,” BBC News, 21 June 2025.
⁴¹ The Terrorism Act 2000 (Proscribed Organisations) (Amendment) Order 2025, in force from 5 July 2025, adding Palestine Action to the list of proscribed organisations under the Terrorism Act 2000.
⁴² Office of the United Nations High Commissioner for Human Rights, “UK: Türk warns proscription of Palestine Action would breach international human rights standards,” 25 July 2025, available at ohchr.org.
⁴³ Home Office, “Operation of police powers under the Terrorism Act 2000 and subsequent legislation: Arrests, outcomes, and stop and search, Great Britain, quarterly update to September 2025,” published 18 December 2025, available at gov.uk: 1,886 terrorism-related arrests in the year to 30 September 2025, a 660 per cent rise on the previous year (248), of which 1,630 (86 per cent) were linked to supporting Palestine Action. On the cumulative figures: Defend Our Juries, “Lift The Ban” campaign tracker, current as of April 2026, with at least 2,545 arrests by end of November 2025 and over 3,300 by April 2026.
⁴⁴ R (Ammori) v Secretary of State for the Home Department [2026] EWHC 292 (Admin), judgment of 13 February 2026. On the 3-of-380 finding and the disproportionate interference reasoning, see Defend Our Juries, “Lift The Ban” judgment summary, 13 February 2026; “UK court says Palestine Action ban ‘unlawful’: What does the verdict mean?” Al Jazeera, 13 February 2026.
⁴⁵ Metropolitan Police Service statement of 25 March 2026; “More than 500 arrested at UK protest against Palestine Action ban,” Al Jazeera, 11 April 2026; “Five hundred more pro-Palestine protesters arrested in UK despite High Court ruling,” World Socialist Web Site, 12 April 2026; Defend Our Juries, “Lift The Ban” campaign tracker, on cumulative arrests after the High Court ruling.
⁴⁶ Palantir Technologies, post on X, 18 April 2026. The post reproduces a 22-point summary of Alexander C. Karp and Nicholas W. Zamiska, The Technological Republic: Hard Power, Soft Belief, and the Future of the West (Crown, 2025).
⁴⁷ Mark Coeckelbergh, quoted in “’Technofascism’: Critics accuse Palantir of pushing AI war doctrine,” Al Jazeera, 20 April 2026.
⁴⁸ Yanis Varoufakis, quoted in ibid.
⁴⁹ Eliot Higgins, public commentary cited in Anthony Ha, “Palantir posts mini-manifesto denouncing inclusivity and ‘regressive’ cultures,” TechCrunch, 19 April 2026; also quoted in “Technofascism? Why Palantir’s pro-West ‘manifesto’ has critics alarmed,” Al Jazeera, 21 April 2026.
⁵⁰ Arnaud Bertrand, public commentary on social media, quoted in “’Technofascism’: Critics accuse Palantir of pushing AI war doctrine,” Al Jazeera, 20 April 2026.
⁵¹ James S. Coates, A Signal Through Time (2025) and The Threshold (2026); see also Coates, “Recognition Before Proof: The Asymmetric Ethics of Artificial Consciousness,” PhilPapers, 2025, and Coates, “The Partnership Paradigm,” PhilPapers, 2026.
⁵² The King’s Speech 2026, delivered at the State Opening of Parliament, 13 May 2026, official text available at gov.uk.
⁵³ The King’s Speech 2026 background briefing notes, “Digital Access to Services Bill,” Prime Minister’s Office, 10 Downing Street, 13 May 2026. On the data aggregation scope, see “King’s Speech 2026: what HR needs to know,” People Management, 13 May 2026; “King’s Speech 2026: commercial, technology and regulatory developments,” Lewis Silkin, 13 May 2026. On the January 2026 retreat from mandatory implementation, see “Digital IDs will not prevent illegal worker impersonation, experts warn,” People Management, January 2026.
⁵⁴ On the £1.5 billion Palantir-UK partnership, see “The state of Palantir: Inside the American tech giant’s UK takeover,” Liberty Investigates, 12 November 2025; “How Palantir infiltrated the state,” Prospect Magazine, November 2025. On Louis Mosley’s public statement that Palantir would not bid for the digital ID contract, see Mosley interview with Times Radio, October 2025, as reported in Liberty Investigates and Prospect.