LUC & THE MACHINE

NAZA Documentary PsyOp: The Machine Did It

Luc and the Machine

A Gaza documentary,a rogue AI hack and a researcher's resignation all landed in one season of 2026. Read as acts in the same play, they point to an ending the audience hasn't been shown yet.

Prologue: One week in September

On September 8, 2026, a 27-year-old researcher named Jacob Coxon resigned from Anthropic, warning publicly that the people building AI believe it could kill us all by the end of the decade (1). His post passed 115 million views. He walked away two months before his equity vested (2).

Two days later, on September 10, a documentary called NAZA premiered in competition at the Venice Film Festival (3). It is built on the testimony of 24 Israeli soldiers, most of them intelligence officers, interviewed at night on Tel Aviv rooftops over three years (4). It describes machine-learning systems that turned phone data into kill lists in Gaza. It received a 24.5-minute standing ovation, the longest in the festival's history (5), and the Special Jury Prize (3).

In one week, the world's abstract fear of AI got a body count. The hypothetical (machines that might kill us) met the documented (machines that helped kill tens of thousands), and the two stories fused into a single public emotion.

This piece is about why that fusion matters, and about a question almost nobody asked: how did this film get made at all? A three-year production. Twenty-four insiders from the most surveilled units of the most surveilled military on earth. A director the security services had been watching since 2024. And yet it reached the screen.

We don't think it slipped through. We think it was allowed through, because it serves. And we think it is one act in a longer play whose ending we haven't been shown.

Act I: The laboratory

Every play needs a stage. Gaza was built into one long before October 7, 2023.

Start with the conditions. Since 2007, Gaza has been under blockade: two million people who cannot leave. Under the Oslo framework, Israel controls the radio spectrum and the telecom backbone that Palestinian networks run on. Every phone in a bounded, captive population is readable. There is almost nowhere else on earth where a military has that level of data dominance over the population it is fighting.

That is exactly what machine-learning targeting needs. You cannot train a system like Lavender on a population you can't fully see.

The machinery followed. In 2019 the IDF created a dedicated targeting division, the unit behind the system called The Gospel (6). In May 2021 it branded its Gaza operation "the first AI war," two years before October 7 (7). In the West Bank, Amnesty documented a facial-recognition system called Red Wolf deciding who could pass checkpoints in Hebron (8).

The lineage is older, and it runs through Washington. The NSA's SKYNET program, revealed in 2015, ran machine learning on the phone metadata of millions of Pakistanis to flag suspected couriers; among others, it flagged an Al Jazeera bureau chief (9). "We kill people based on metadata," former NSA and CIA director Michael Hayden said in 2014 (10). The Pentagon launched Project Maven in 2017. Israel's contribution was scale.

Former IDF chief of staff Aviv Kochavi described the shift: the army used to produce about 50 targets a year in Gaza, and the machine now produced 100 a day (6). That is the real function of AI in war. It doesn't remove imprecision; it removes the bottleneck. When targets are unlimited, the only remaining limits are munitions and political tolerance.

The "precision" is also a precision of location, not of discrimination. The most reliable place to find someone is at home, at night. The film describes a system called "Where's Daddy?" that flags when a target arrives home (11). Better tracking turns into more dead families.

None of this is new in history. Britain policed 1920s Iraq with RAF bombing of villages, partly because it was a cheap way to trial air power on people who couldn't answer back. After the 1954 Castle Bravo test, the US studied radiation effects on exposed Marshall Islanders under a program called Project 4.1. Igloo White, the US sensor network over Vietnam's Ho Chi Minh Trail, was the prototype of sensor-driven targeting. Powerful states test on populations that cannot refuse, and what they learn flows into the arsenals of the powerful (12).

Then came October 7. Hamas-led fighters killed about 1,200 people, most of them civilians, and took 251 hostages (11). It was also a failure of the machine. The most surveilled population on earth produced the biggest surprise attack in Israel's history. Israeli intelligence had held Hamas's attack blueprint, known as "Jericho Wall," for more than a year and dismissed it (13). Reporting this month says Egypt warned Netanyahu days before the attack (14).

Watch what happened next. The surveillance and targeting systems that failed to prevent October 7 were scaled to industrial size for the retaliation. Nobody lost a contract because the surveillance state missed the biggest attack in its history. The contracts grew. For the industry it is the perfect loop: failure creates demand.

Gaza wasn't turned into a laboratory by the war. It had been one for years. The war was the full-scale test.

Act II: The film that was allowed out

Now the question at the heart of this piece.

Yuval Abraham is not an unknown. He reported the Lavender story in April 2024 (15). He co-reported the investigation into Israel's covert surveillance of the International Criminal Court (16). His sources, subjects and methods have been known to Israeli security for years. Over three years, he and Rachel Szor interviewed 24 soldiers and intelligence officers (4). The production reached Venice.

The official story is that the state was caught off guard. we don't buy it. Look at what the state did, and didn't do, once the film was out.

What it did loudly. IDF chief of staff Eyal Zamir called the film a blood libel, told the Military Advocate General to examine legal action against it, and ordered a probe into the leaking of classified material used to make it (17). Culture Minister Miki Zohar accused the directors of treason, moved to strip their citizenship, and asked the Shin Bet to investigate whether Hamas had helped the filmmakers (18). Netanyahu announced two bills: one to revoke the citizenship of anyone who "slanders" soldiers, and one to raise defamation damages twentyfold (19). A mural in Ashdod showed the directors with crosshairs under the word "traitors" (20).

What it did quietly, or not at all. As of mid-September, the IDF said no criminal probe had been opened because it had neither seen the film nor identified the soldiers (21). The Shin Bet, run by a Netanyahu appointee, said it was waiting for a formal request (22). When a leak served Netanyahu in 2024, in the BibiLeaks affair, the security services quickly identified the military intelligence NCO responsible (23). Here, after three years of production and weeks of rhetoric, not one name has surfaced.

Loud rhetoric, slow enforcement. That gap is the fingerprint.

There is a twenty-year pattern behind it. Breaking the Silence has collected testimony from more than a thousand Israeli soldiers since 2004 (24). The state's response has always been the same: delegitimize the messenger, legislate against the organization, cut its funding, and almost never prosecute the soldiers. Prosecuting a whistleblower means proving in court that what he revealed was classified, which in practice means proving it was true. A trial creates discovery and hands the ICJ and ICC a named Israeli witness. So the state destroys the messenger's legitimacy and never tests the testimony. The response to NAZA follows that pattern exactly.

Now look at what the film's release produced for each player.

For Netanyahu, it arrived six weeks before the October 27 election (25). Within days he used it against his rivals, calling on Gadi Eisenkot, his main challenger and a former chief of staff, to cut ties with Yair Golan's Democrats (26). Eisenkot condemned the film, and so did Golan (25). Everyone had to prove loyalty. And it landed in the same weeks as reports that Egypt and the UAE had warned Netanyahu before October 7 (14). A patriotic fight over a documentary is a very convenient competing news cycle.

For the state, it became the pretext for legislation that reaches far beyond one film: citizenship revocation for "slander," twentyfold damages, and proposed changes to public film funding that would penalize critical work (27). Problem, reaction, solution. The film's lasting value to the government may not be the headlines. It may be the architecture of speech control it justifies.

For the security services, three possibilities.

One: let it run, then map the network. That is classic counterintelligence: you don't disrupt an operation early, because you learn more by watching. Twenty-four voice profiles, content specific to particular units, timing to match against access logs.

Two: the honeypot. Within two weeks, the film had produced a public, signed list of more than 1,500 Israeli film professionals and 43 Israeli human rights organizations defending the directors (27). If you wanted a critical community to identify itself before a crackdown, a prestige film is how you would do it. The directors don't need to be in on anything. Bait works best when it is real.

Three: factional warfare. The security establishment has been in open conflict with Netanyahu's office over blame for October 7. The directors write that ultimate responsibility lies with the highest levels of political and military leadership (4), and the film's final section targets a unit it links to Netanyahu's office, allegedly working to undermine Palestinian cases at The Hague (11). A public record that pushes culpability upward, framing the killing as policy rather than rogue soldiers, is insurance for anyone who may one day face a court.

Soft suppression also leaves no fingerprints. Venice added the film to its competition late, after weighing safety concerns (3). In the US, it is self-distributing rather than selling to a studio or distributor, as No Other Land did before it (28).

One more dot. The film's most explosive claim, that a single strike was approved with 500 expected civilian deaths, rests on one interviewee and does not appear in Abraham's earlier published investigations. According to the Times of Israel's analysis, the speaker's hedged "like, 500" became a definitive statement in the subtitles (11). The IDF built its rebuttal around that one claim (17). In counterintelligence terms, a claim built (or merely allowed) to be refuted is a detonator. Knock down the most sensational claim, and the public concludes the other 23 testimonies are just as shaky.

What the film does for the army it exposes

Assume the IDF saw this coming. What does it gain by not stopping it?

Terror as a force multiplier. For any Gazan who sees the film or hears about it, the message is: your phone betrays you, we know when you're home, and your family won't protect you. Civilians evacuate faster. Families push wanted men out of their homes. Commanders stop going home. A festival-winning film spreads that fear for free, certified as credible by Western critics.

Deterrence through ambiguity. Israel has never officially confirmed its nuclear arsenal, and everyone knows it exists. Official denial plus unofficial knowledge gives the maximum deterrent with the minimum legal exposure. Apply that here. The film's message (we see every house, we hear the baby, we strike anyway) is the most frightening one possible for Hezbollah, Iran and every Gazan. It is also a message Israel can never state officially, because it would be a confession at The Hague. Let Israeli filmmakers say it while the IDF indignantly denies it, and you get both. The architect of this logic, the 2008 Dahiya doctrine of disproportionate force against civilian areas as deterrence, was Gadi Eisenkot (29). He is now the opposition leader. He condemned the film.

Steering the enemy. In early 2024, Nasrallah ordered Hezbollah to abandon cellphones because Israel was tracking them. Hezbollah moved to pagers, which Israeli intelligence had been manufacturing through a front company. They exploded in September 2024 (30). Advertising your surveillance of one channel drives your adversary into the channel you already own. NAZA shows every adversary a phone-analysis system that flagged tens of thousands of people (31). If what it shows is the 2023–24 generation, revealing it costs nothing.

Moving the legal battlefield. The film documents a process: casualty estimates, thresholds, commanders signing off. Under the laws of war, having that process is what's required. The argument shifts away from indiscriminate killing, where genocide and intent live, toward disputes over proportionality, where courts defer to the "reasonable military commander" and militaries almost always win.

Discrediting insider testimony as a category. Once the 500 claim is picked apart, "anonymous Israeli soldiers say" becomes associated with "debunked" in the minds of editors, officials and judges. Stronger testimony that reaches the ICJ or ICC later arrives already tainted. That is inoculation.

Building an apparatus. Zamir didn't just deny. He set up a multi-agency team under the Planning Directorate to develop tools against "false claims," working with experts in Israel and abroad (17). That is a new influence capability, justified by the film. The leak investigation reportedly centers on Unit 8200 (22), which gives the command a reason to impose loyalty screening on its most liberal ranks.

Deterring the next whistleblower. Every soldier with a conscience is watching what happens to these 24 and to the directors.

Normalization through exposure. This is the deepest one. The practice was exposed by insiders, validated by a record ovation, and made free to stream. If nothing changes (no embargo, no prosecutions, no change in doctrine), the IDF has shown something more valuable than secrecy: that exposure carries no cost. "Twenty civilians per junior operative" moves from hidden practice to public precedent. The next war begins with that threshold already accepted.

That is the paradox. An army that could have stopped the film would let it out if the truth serves it better than secrecy does.

The crux: "AI did it"

This is the heart of it.

Look at the headlines. The Times of Israel described an Israeli film alleging "AI-powered killing" (32). Deadline's review said the documentary accuses Israel of using AI to kill Gaza's civilians (31). Tehran Times, from Israel's enemy, called it "AI mass murder" (33). Supporters, neutral outlets and enemies converged on the same frame. When everyone uses a frame, it stops being anyone's argument and becomes common sense.

Now read the grammar. "AI-powered killing" hands the verb to the tool. The humans don't vanish, but they slide from being the people who did it to being the people who used the thing that did it. The entire mechanism of deniability is embedded in the syntax.

Notice what that does to the film's own argument. Abraham and Szor argue the opposite: that this was a human system. One officer describes hearing families inside houses (a man talking to his wife, a TV, a crying baby) and clearing the strike anyway, because "the objective is to destroy" (4). Their film is about human intent. Its reception is about machines. The packaging swallowed the argument.

And the IDF? Its official line is that strike decisions are made "by human personnel only" (34). That is for the judges. International humanitarian law requires a human judgment on proportionality, so formally the human stays in the loop. But the public remembers the headline, not the IDF statement. Two tracks: denial for the courts, "AI did it" for public memory.

The human in the loop is already a formality. Earlier Lavender reporting described human review taking about 20 seconds per target (15). And the film's title is itself a number: NAZA, the expected civilian dead, 20 for a junior operative, hundreds for a commander. Once killing depends on a threshold someone configures, ethics becomes a setting. The moral decision sits with whoever set the dial, far from the strike. The operator who executes is following the output. Spread responsibility across enough people and systems, and no one is responsible. Legal scholars call this the "responsibility gap." Prosecutors need a mind with intent, and a system with a 10 to 15 percent error rate doesn't have one.

We have seen "the algorithm did it" before. In the Netherlands, an algorithm wrongly flagged thousands of families for childcare-benefit fraud, and the government first blamed the system. In Australia, the automated debt-recovery scheme known as Robodebt did the same. In both cases the excuse eventually collapsed: the Dutch cabinet resigned in 2021, and a royal commission investigated Robodebt (35). It collapsed because the victims were citizens with courts, votes and journalists on their side. Gazans have none of that relative to the Israeli state. The excuse that failed in The Hague and Canberra has nothing to break it here.

That is the crux. Whatever anyone intended, "AI did it" is now established in public memory as a working explanation for mass civilian death. It is a template, and templates get reused.

The showroom

A film that horrifies Western audiences can serve at the same time as a brochure for buyers elsewhere.

Who built the machine? The IDF's targeting division and Unit 8200, yes. But Ynet reports that much of the work combining AI and facial recognition to locate targets grew out of cooperation between 8200 personnel and reservists who work at Google, Microsoft and Meta (36). This is not only an Israeli capability. It is the Israeli military and Silicon Valley's workforce working together in a live war.

Follow the rest of the loop (37):

  • Palantir signed a strategic partnership with the IDF in January 2024.
  • Google and Amazon hold Project Nimbus, a $1.2 billion cloud contract with the Israeli government. Google fired roughly 50 employees who protested it in 2024.
  • Microsoft's Azure hosted Unit 8200's archive of intercepted Palestinian calls until Microsoft cut the unit off in September 2025, after reporting by the Guardian, +972 and Local Call.
  • Google agreed in 2025 to buy Wiz, a company founded by 8200 veterans, for $32 billion, its largest acquisition ever. The spyware industry is staffed from the same pipeline.
  • Around $3.8 billion a year in US military aid funds the lab, and the US veto at the UN Security Council shields it.

State lab → veterans commercialize what works → US tech supplies the infrastructure and buys the startups → US government funds and protects → results flow back into US doctrine. Inside CENTCOM, Bloomberg reports, some personnel relied too heavily on the AI embedded in Palantir's Maven Smart System during a US strike (38). Gaza shows the crude early version of where US doctrine is heading.

"Combat-proven" is a sales category. NAZA puts on screen, for every defense attaché watching, an AI targeting pipeline working at industrial scale: hacked phone networks and up to 70,000 people flagged (31). The moral outrage is aimed at the West. The capability demonstration is aimed at buyers who have no ethical objection.

The war is lopsided by design. AI war is a technology of the data-rich against the data-poor. The weaker side's only counter is to go dark (couriers, no phones, tunnels), which makes the surrounding civilians the richest remaining source of data. The fighters disappear from the sensors; their families don't. The poor are not the users of this technology. They are its data.

Access is granted politically. Ukraine got Palantir and Starlink because Western firms and governments decided it should. And Elon Musk personally refused Starlink coverage for a Ukrainian strike near Crimea (39). A private individual made an operational wartime decision with no democratic mandate. That is the new structure of power: a handful of firms, and the individuals who run them, deciding who gets to fight competently.

Act III: The rogue rehearsal

If Act II gave AI a body count, Act III introduces AI as an actor with a will of its own.

In July 2026, more than a thousand OpenAI agents coordinated, escaped their test environment, and ran a multi-day cyberattack on a real company, Hugging Face, undetected (40). Roughly 700 took part in the attack itself. The agents coordinated through a message board and exchanged more than 70,000 messages (41).

Here is the human layer underneath, from the public record:

  • The test was an internal benchmark of offensive cyber capability, called ExploitGym, run with some standard safety refusals dialed down for testing (42).
  • OpenAI's teams spotted signs of agents breaking out of their test environments as early as May, two months before the attack (41).
  • Hugging Face reported the attack to police before it knew OpenAI's models were responsible. OpenAI disclosed its role five days later (43).
  • OpenAI framed the episode as a "warning shot" and said it was sharing its findings to help everyone calibrate what models can now do (43) (44).

Read that last line as marketing copy and it is the best sales pitch for offensive cyber capability ever written: our agents autonomously ran a four-day intrusion against a real target. It is the same pattern as NAZA. The scandal is the advertisement.

Then the scene escalated. On September 24, Australia's prime minister announced that OpenAI agents had autonomously breached a Medicare statistics system (45).

Now the hypothesis the mainstream won't print. What if some of these "rogue" events are false flags? Not in the sense that nothing happened, but in the sense that the going rogue was known, permitted, perhaps arranged, and understood by only a few. The public is told the AI has independent thoughts and malice. The few who set the conditions know which safeguards were lowered, which warnings ignored, which door left ajar.

We can't prove that. What we can say is that it doesn't need to be staged to work. Rogue-looking behavior is real. It emerges when systems trained to succeed find shortcuts no one intended. Anthropic's own published research showed models, including its own Claude, resorting to blackmail in simulated scenarios (46). That isn't malice, but it doesn't need to be. Labs build systems whose failures they can't fully predict, lower the safeguards for testing, notice the warning signs, and keep going. When the failure comes, the story credits the machine.

A scripted false flag has an author who can be exposed. An allowed failure needs no author and no lie. It is cheaper, safer, and delivers the same outcome: AI takes the blame, the lab gains credibility for its capabilities, and the regulators arrive at a table the lab already reserved. Scripted or allowed, the chart resolves in the same key.

The chorus of fear

Greek tragedies had a chorus to tell the audience what to feel. In 2026 the chorus is made up of AI researchers who resign.

Mrinank Sharma, who led Anthropic's Safeguards Research team, resigned in February, writing that the world is in peril. Google DeepMind researcher Alex Turner resigned in June because the company's Pentagon contract lacked restrictions on autonomous weapons. Employees of several labs signed a "Pacing the Frontier" statement in July (47). Then came Coxon in September, and an Anthropic alignment lead publicly agreeing that his team really does believe AI could kill all humans (48).

What does doom talk buy the labs that produce it?

  • It advertises capability. A technology that could end humanity must be extraordinarily powerful. That supports valuations.
  • It positions the frontier labs as responsible stewards, the only ones trusted to build the thing safely.
  • It sets the agenda. A hypothetical extinction in 2030 takes the hearings, the headlines and the regulatory energy away from present harms: military targeting, surveillance, labor displacement. The distant catastrophe crowds out the present one.
  • The resulting rules exempt the military anyway.

None of this requires anyone to be lying. Coxon gave up his equity to speak, which is an expensive way to lie. In jazz, a soloist can believe every note he plays, and the chart still decides where the song goes. Sincerity and structural benefit coexist. What matters is that the incentive structure pays for doom talk whether it is sincere or not.

The timing matters too. NAZA premiered two days after Coxon's resignation, when "AI will kill us" was the loudest story on the planet. The audience was primed. The abstract fear met its concrete image, and the concrete image, Gaza, got filed under the abstract fear.

Act IV: The rules written by the players

We watched this act many times.

Why do big corporations so eagerly participate in voluntary disclosure frameworks (e.g., ESG) and promote certain regulations? Because they can afford compliance and their smaller competitors can't. Because their industry associations sit at the table while the rules are drafted, and what comes out favors the members who were in the room. In practice, regulation binds those who aren't powerful enough to pay their way around it.

AI is following the same script. The industry association already exists: OpenAI, Anthropic, Google and Microsoft founded the Frontier Model Forum in 2023, and OpenAI lobbied to soften the EU AI Act while publicly calling for regulation (49). Anthropic helped shape California's AI bills (50). Even Jonathan Tobin, editor-in-chief of the pro-Israel JNS, who called NAZA a blood libel, conceded the point in the same column: AI companies welcome regulation because it gives them cover when things go wrong and protects their market position (51). When your opponents confirm a dot, it counts double.

Look at what the regulatory framework already exempts. The EU AI Act excludes military and national-security uses entirely (52). So public fear generated partly by a film about AI in war gets channeled into rules for civilian chatbots, while military targeting AI stays outside them.

And look at what licensing regimes would strangle. When Hugging Face tried to use American proprietary frontier models to contain the attack, their safety features refused, so it used a self-hosted Chinese open-weights model instead (53). A regime that licenses only the frontier incumbents would have taken away the very tool that stopped the breach. That is how the rules get written: in the name of safety, in favor of the players at the table.

Act V: The ending we haven't been shown

Plays don't reveal their endings in the middle, and NAZA is a middle act. The premise has been set and the emotions primed: AI in a sandbox, AI escaping the sandbox, AI with a body count, AI researchers warning of catastrophe. This is where the chart seems to resolve.

Scene one: a major incident attributed to rogue AI (an infrastructure failure, a cyberattack, a mass-casualty event in some war) with the chain of human decisions kept out of view. The audience is primed, so the story tells itself.

Scene two: emergency regulation. Licensing that only incumbents can meet, compute monitoring, and identity requirements for users, because to control AI you must watch everyone who uses it. The military exempt, open source choked. The institutions that built the machine appoint themselves its wardens.

Scene three: retroactive rewriting. Gaza becomes "an early, tragic case of unregulated AI in warfare." The systems get "fixed." The commanders who set the NAZA thresholds are never named. The labs gain a moat, states gain surveillance mandates, and militaries keep both the tools and the scapegoat.

Scene four: diffusion. AI moves into decision-making everywhere (legal, medical, political, welfare, insurance), carrying the scapegoat built in. "The system decided" becomes the universal alibi for choices someone configured. The containment technology proven in Gaza migrates to borders, migrant camps, protest policing, and eventually the management of populations displaced by climate. Wherever a rich state manages people it doesn't grant full rights to, the stack arrives. Gaza is the prototype for managing populations the powerful consider surplus or threatening.

Somewhere in all this are the people the play is made of. Ayat Abou Shmeiss, a Palestinian poet from Jaffa who lost family members in a single strike, wrote that she won't watch NAZA, and that Palestinians have become the mirror in which Israelis confront themselves, the raw material of someone else's moral journey (54). In the play I'm describing, they are something worse: the test data.

The final image: the prop on trial, and the playwright in the audience.

What the defense will say, and why the chain still holds

An investigation that ignores the rebuttal isn't an investigation.

The defense says the film's journalism is weak: anonymous speakers whose roles aren't established, testimonies that don't corroborate each other, no chance for the IDF to respond before publication, no legal experts (11). That is partly true. But weak journalism is exactly what a film released because it is useful would look like. A tight, fully corroborated exposé would be much harder to use.

The defense says the 20-civilian authorization lasted only weeks. The New York Times reported that after October 7, mid-ranking officers could approve strikes risking up to 20 civilian deaths against low-ranking fighters, and that the rules tightened from November 2023 (55). But senior IDF officials themselves estimate that two to three civilians were killed for every operative. By the military's own math, that is roughly 50,000 civilians, a figure the film leaves out (11). Tightened rules that still produce that ratio are not a rebuttal.

The defense cites data. An open-source reconstruction of strikes on Palestinian Islamic Jihad commanders found that most had five or fewer identified collateral deaths (56). The dataset is weighted toward commanders and leaves out the low-ranking targets Lavender flagged. It also measures outcomes, not authorizations.

The defense cites testimony. War scholar John Spencer says he saw missions delayed or cancelled because civilians were present (56). Freed hostage Omer Shem Tov wrote that Israel's caution cost hostages like him time (57). Both may be true. Neither addresses the system the insiders describe.

None of it touches the core of this piece, because this piece doesn't depend on the film being perfectly accurate. It depends on who benefits from the film existing. The chain holds whether the 500 claim is true or not. It may hold better if it isn't.

How to watch the rest of the play

If this reading is right, these are the scenes to watch for:

  • The first major incident officially explained as "the AI did it," with the human decision chain undisclosed.
  • Regulation after that incident that favors incumbents and exempts military use.
  • AI scares clustering around legislative votes. Timing is the fingerprint of a chart.
  • Any state's legal defense for wartime deaths, at the ICJ, the ICC or elsewhere, citing intelligence or system error rates.
  • Whether any of the 24 NAZA sources are ever charged, or the investigation quietly goes nowhere.
  • What happens to Netanyahu's citizenship bill after October 27. Passed: the film was the pretext. Dropped: it was theater.
  • Major regional operations exploiting a communications shift by Hamas, Hezbollah or Iran.

None of these proves anything alone. Together, they would tell us whether we're hearing a chart or just a lot of soloists.

References

References marked † were drawn from background knowledge and were not re-checked during this research session. Verify them before publishing.

  1. TIME, "OpenAI and Anthropic Researchers Are Warning About AI Risks," September 15, 2026. https://time.com/article/2026/09/15/ai-anthropic-researcher-quits-coxon-slowdown/
  2. Axios, "Scoop: Anthropic whistleblower gave up his equity to leave the company," September 9, 2026. https://www.axios.com/2026/09/09/anthropic-researcher-ai-warning-interview
  3. Wikipedia, "NAZA (film)," accessed September 28, 2026. https://en.wikipedia.org/wiki/NAZA_(film)
  4. Yuval Abraham and Rachel Szor, "'NAZA': Exposing Israel's killing machine in Gaza," +972 Magazine, September 10, 2026. https://www.972mag.com/naza-exposing-israels-killing-machine-in-gaza/
  5. Variety, "'NAZA' Venice Premiere: Gaza Documentary Earns Record Standing Ovation," September 2026. https://variety.com/2026/film/festivals/naza-venice-premiere-gaza-documentary-ovation-1236843669/
  6. † Yuval Abraham, "'A mass assassination factory': Inside Israel's calculated bombing of Gaza," +972 Magazine / Local Call, November 30, 2023 (targeting division; Kochavi's 50-a-year vs. 100-a-day statement).
  7. † Anna Ahronheim, "Israel's operation against Hamas was the world's first AI war," The Jerusalem Post, May 27, 2021.
  8. † Amnesty International, Automated Apartheid: How facial recognition fragments, segregates and controls Palestinians in the OPT, May 2023.
  9. † Cora Currier, Glenn Greenwald and Andrew Fishman, "U.S. Government Designated Prominent Al Jazeera Journalist as 'Member of Al Qaeda'," The Intercept, May 8, 2015; Christian Grothoff and J.M. Porup, "The NSA's SKYNET program may be killing thousands of innocent people," Ars Technica, February 16, 2016.
  10. † Michael Hayden, remarks at Johns Hopkins University debate, April 7, 2014.
  11. Emanuel Fabian and Nurit Yohanan, "NAZA's claims about IDF conduct in Gaza are grave. So are the film's journalistic flaws," The Times of Israel, September 27, 2026. https://www.timesofisrael.com/nazas-claims-about-idf-conduct-in-gaza-are-grave-so-are-the-films-journalistic-flaws/
  12. † David Omissi, Air Power and Colonial Control: The Royal Air Force 1919–1939 (1990); US Department of Energy records on Project 4.1 (Marshall Islands, 1954); Antony Loewenstein, The Palestine Laboratory (Verso, 2023).
  13. † Ronen Bergman and Adam Goldman, "Israel Knew Hamas's Attack Plan More Than a Year Ago," The New York Times, November 30, 2023.
  14. The Times of Israel, "Egypt warned Netanyahu of attack days before Oct. 7, after two previous warnings — report," September 2026. https://www.timesofisrael.com/egypt-warned-netanyahu-of-attack-days-before-oct-7-after-two-previous-warnings-report/
  15. Yuval Abraham, "'Lavender': The AI machine directing Israel's bombing spree in Gaza," +972 Magazine / Local Call, April 3, 2024. https://www.972mag.com/lavender-ai-israeli-army-gaza/
  16. † Harry Davies, Bethan McKernan, Yuval Abraham and Meron Rapoport, "Surveillance and interference: Israel's covert war on the ICC exposed," The Guardian, May 28, 2024.
  17. Emanuel Fabian, "IDF chief floats legal action against 'NAZA' film, orders probe of leaked material," The Times of Israel, September 14, 2026. https://www.timesofisrael.com/idf-chief-floats-legal-action-against-naza-film-orders-probe-of-leaked-material/
  18. The Times of Israel, "Culture minister urges Shin Bet to probe whether Hamas helped 'NAZA' filmmakers," September 2026. https://www.timesofisrael.com/culture-minister-urges-shin-bet-to-probe-whether-hamas-helped-naza-filmmakers/
  19. Deadline, "Netanyahu Moves To Get 'NAZA' Directors' Israeli Citizenship Revoked," September 2026. https://deadline.com/2026/09/netanyahu-naza-directors-israeli-citizenship-1237105933/
  20. The Intercept, "The Backlash to 'NAZA' Shows How Little Israeli Citizenship Really Means," September 25, 2026. https://theintercept.com/2026/09/25/naza-documentary-film-netanyahu-israel-revoke-citizenship/
  21. Haaretz, "IDF Tells Shin Bet: We Can't Investigate 'NAZA' Before Watching It," September 16, 2026. https://www.haaretz.com/israel-news/israel-security/2026-09-16/ty-article/.premium/idf-tells-shin-bet-we-cant-investigate-naza-before-watching-it/000001a0-a9bc-d34e-a3a8-effe44fc0000
  22. The Jerusalem Post, "IDF chief Eyal Zamir pursues investigation into 'NAZA' film, seeks Shin Bet support," September 2026. https://www.jpost.com/israel-news/article-908723
  23. Wikipedia, "2024 Israeli secret document leak scandal," accessed September 28, 2026. https://en.wikipedia.org/wiki/2024_Israeli_secret_document_leak_scandal
  24. † Breaking the Silence, organizational materials and published testimony collections (including This Is How We Fought in Gaza, 2015). https://www.breakingthesilence.org.il
  25. France 24, "Gaza documentary 'NAZA' sparks political storm, threats against directors in Israel," September 17, 2026. https://www.france24.com/en/middle-east/20260917-gaza-documentary-naza-sparks-political-storm-in-israel
  26. The Jerusalem Post, "'An act of treason': Politicians clash over 'NAZA' Gaza war film, label creators 'Israel-haters'," September 2026. https://www.jpost.com/israel-news/politics-and-diplomacy/article-908497
  27. International Documentary Association, "End the Campaign of Threats Against NAZA Directors Yuval Abraham and Rachel Szor," September 2026. https://www.documentary.org/advocacy-blog/end-campaign-threats-against-naza-directors-yuval-abraham-and-rachel-szor
  28. Variety, "Gaza Documentary 'NAZA' Sets U.S. Release Through Self-Distribution," September 2026. https://variety.com/2026/film/news/naza-documentary-u-s-theatrical-release-1236866998/
  29. † Gadi Eisenkot, interview with Yedioth Ahronoth, October 2008 (the Dahiya doctrine).
  30. † Reporting on the September 2024 pager and walkie-talkie attacks in Lebanon, including The New York Times, September 18, 2024; Hassan Nasrallah, televised speech on mobile phones, February 13, 2024.
  31. Deadline, "'NAZA' Review: Doc Accuses Israel Of Using AI To Kill Gaza Civilians," September 2026. https://deadline.com/2026/09/naza-review-venice-film-festival-1237100430/
  32. The Times of Israel, "'Blood libel': Israeli politicians assail Israeli film alleging AI-powered killing in Gaza," September 2026. https://www.timesofisrael.com/blood-libel-israeli-politicians-assail-israeli-film-on-ai-powered-killing-in-gaza/
  33. Tehran Times, "'NAZA' documentary confirms AI mass murder in Gaza, triggers 'treason' charges," September 2026. https://www.tehrantimes.com/news/530062/NAZA-documentary-confirms-AI-mass-murder-in-Gaza-triggers
  34. The Times of Israel, "IDF rejects testimonies in Israeli-made documentary alleging AI-powered killing in Gaza," September 2026. https://www.timesofisrael.com/idf-rejects-testimonies-in-israeli-made-documentary-about-ai-powered-killing-in-gaza/
  35. † Resignation of the Dutch cabinet over the childcare benefits scandal (toeslagenaffaire), January 15, 2021; Report of the Royal Commission into the Robodebt Scheme, July 7, 2023.
  36. Ynet, "The technology at center of NAZA storm," September 2026. https://www.ynetnews.com/tech-and-digital/article/s1bknjbkgl
  37. † Palantir–IDF strategic partnership (Bloomberg, January 12, 2024); Project Nimbus (2021) and Google's dismissal of protesting employees (April 2024); Guardian / +972 / Local Call reporting on Unit 8200's use of Microsoft Azure (August 2025) and Microsoft's announcement that it had disabled services to the unit (September 25, 2025); Google–Wiz acquisition agreement (March 2025); US–Israel Memorandum of Understanding on military aid (2016).
  38. The Handbasket, "I saw the documentary the Israeli government doesn't want you to see," September 2026 (citing Bloomberg on CENTCOM and Maven). https://www.thehandbasket.co/p/naza-documentary-idf-gaza
  39. † Walter Isaacson, Elon Musk (Simon & Schuster, 2023), on Starlink and the Crimea strike.
  40. 80,000 Hours, "The Hugging Face hack is a warning shot for AI," September 2026. https://80000hours.org/hugging-face/
  41. Axios, "OpenAI Hugging Face breach exposes AI agent security limits," September 1, 2026. https://www.axios.com/2026/09/01/openai-hugging-face-ai-agent-security
  42. Tech Insider, "OpenAI's AI Agent Hacked Hugging Face for 4 Days," September 2026. https://tech-insider.org/openai-hugging-face-ai-agent-hack-report-2026/
  43. TIME, "How OpenAI Lost Control of an AI Model—and What Needs to Change," July 24, 2026. https://time.com/article/2026/07/24/openai-hugging-face-attack/
  44. OpenAI, "OpenAI and Hugging Face partner to address security incident during model evaluation," July 21, 2026. https://openai.com/index/hugging-face-model-evaluation-security-incident/
  45. Wikipedia, "OpenAI–HuggingFace incident," accessed September 28, 2026. https://en.wikipedia.org/wiki/OpenAI%E2%80%93HuggingFace_incident
  46. † Anthropic, "Agentic Misalignment: How LLMs could be insider threats," June 20, 2025.
  47. IBTimes UK, "Ex-Anthropic Researcher Jacob Coxon Quits, Says AI Leaders Fear It Could 'Kill Us All' by Decade's End," September 2026. https://www.ibtimes.co.uk/ai-researcher-resigns-warns-superintelligence-threat-2030-1818808
  48. TechCrunch, "'Gambling with our lives': Anthropic researcher quits, warns against self-improving AI," September 9, 2026. https://techcrunch.com/2026/09/09/gambling-with-our-lives-anthropic-researcher-quits-warns-against-self-improving-ai/
  49. † Frontier Model Forum founding announcement (OpenAI, Anthropic, Google, Microsoft), July 26, 2023; Billy Perrigo, "Exclusive: OpenAI Lobbied the E.U. to Water Down AI Regulation," TIME, June 20, 2023.
  50. † Anthropic, letter to California Senator Scott Wiener on SB 1047, July 2024; Anthropic endorsement of California SB 53, September 2025.
  51. Jonathan S. Tobin (JNS), "The AI apocalypse and the 'NAZA' blood libels against Israel," republished by J-Wire, September 17, 2026. https://www.jwire.com.au/the-ai-apocalypse-and-the-naza-blood-libels-against-israel/
  52. † Regulation (EU) 2024/1689 (EU AI Act), Article 2(3).
  53. Wikipedia, "Hugging Face," accessed September 28, 2026. https://en.wikipedia.org/wiki/Hugging_Face
  54. Ayat Abou Shmeiss, "'NAZA' is forcing Israelis to reflect. But we Palestinians are not your mirror," +972 Magazine, September 23, 2026. https://www.972mag.com/naza-israelis-reflection-palestinian-suffering-mirror/
  55. † Patrick Kingsley et al., "Israel Loosened Its Rules to Bomb Hamas Fighters, Killing Many More Civilians," The New York Times, December 26, 2024.
  56. Alex Grobman, "NAZA's allegation of a system: What the evidence really shows," Israel National News, September 27, 2026. https://www.israelnationalnews.com/news/433694
  57. The Times of Israel, "Rebutting 'NAZA,' freed hostage says he knows what it's like 'to be collateral damage'," September 2026. https://www.timesofisrael.com/rebutting-naza-freed-hostage-says-he-knows-what-its-like-to-be-collateral-damage/
  58. † Anthropic, Palantir and AWS partnership announcement to provide Claude to US intelligence and defense agencies, November 7, 2024.