Twenty-five years ago this week, the NGO Forum at the UN's Durban conference accused Israel of apartheid, genocide and ethnic cleansing and called for sanctions, embargoes and international isolation modeled on the campaign against apartheid South Africa. Its
declaration called Israel a “racist, apartheid state,” accused it of genocide and ethnic cleansing, and demanded its “complete and total isolation.”
This was September 2001. There was no blockade of Gaza, no October 7, no Gaza wars of 2008, 2014, 2021 or 2023, no current claims of deliberate starvation in Gaza and no supposed Gaza genocide. Most of the evidence cited today to explain why respectable institutions have concluded that Israel is an apartheid, genocidal, ethnic-cleansing state did not exist yet.
But the conclusion did.
I'm not saying Durban secretly planned the next quarter-century of anti-Israel activism. I am saying something more important: much of the conceptual framework through which Israel is judged today preceded the evidence now used to justify it. The verdict came first. Over the following 25 years, arguments and purported evidence accumulated beneath it, while the framework itself spread through NGOs, academia, UN institutions, international law, journalism and governments until it created something much more powerful than any individual accusation: an epistemic environment in which Israel's guilt increasingly appears to be the only reasonable conclusion.
Fact checks can counter individual lies. But they cannot by themselves stop an epistemic environment built on a tsunami of lies, half-truths, biased reports, circular references, changed definitions, missing denominators and assumptions of guilt used as the basis for further analysis.
When bad evidence becomes accepted evidence
Consider what someone who casually follows the news “knows” about Israel today. Israel practices apartheid and ethnic cleansing. Settlers terrorize Palestinians to drive them from their homes. Israel deliberately starved Gaza and created famine. It systematically destroys hospitals, deliberately kills journalists and aid workers, indiscriminately bombs civilians, illegally occupies Palestinian land, runs illegal settlements, embodies settler colonialism and is committing genocide.
I have spent 22 years examining accusations like these, and in case after case the problem is not merely that genuine evidence is framed unfairly. Often the evidence itself does not survive examination.
Sometimes the source doesn't say what the headline says. Sometimes statistics are presented without a meaningful denominator. Sometimes combatants disappear into civilian casualty figures. Sometimes NGOs infer Israeli intent without access to Israeli intelligence and then treat their inability to identify a military target as proof that no target existed. Sometimes supposedly independent reports repeat the same original allegation. Sometimes inconvenient contrary evidence disappears.
And sometimes the definition of the crime itself changes.
That last part is especially important, because I have been watching it happen for years. “Occupation,” “apartheid” and now “genocide” did not simply have established legal definitions that neutral organizations applied to Israel and found that Israel unfortunately fit. Again and again, major human-rights organizations have stretched, combined or discarded parts of existing definitions until Israel fits the accusation.
Take occupation. In 2003, when Amnesty International explained belligerent occupation in Iraq, it used the traditional Hague standard: territory is occupied when it is actually under the authority of the hostile army, with foreign forces exercising effective control on the ground, and occupation ends when those forces relinquish that control. Yet years after Israel removed its permanent troops and settlements from Gaza and Hamas exercised governmental control there, Amnesty continued calling Israel the “occupying power.” I pointed out the contradiction back in 2010: Amnesty's own definition of occupation for Iraq contradicted the definition it was applying to Gaza.
Then came apartheid. The Apartheid Convention and Rome Statute define apartheid in terms of domination by one racial group over another. Human Rights Watch needed a broader concept to make its case against Israel, so it borrowed the expansive definition of “racial discrimination” from a different treaty, the International Convention on the Elimination of All Forms of Racial Discrimination. But, as I pointed out, HRW did not quote the very next paragraph of that convention, which says the convention does not apply to distinctions between citizens and non-citizens—the distinction on which a huge portion of the Israel-apartheid argument depends. HRW borrowed the part of ICERD it needed and ignored the paragraph that cut against its argument.
Amnesty then went further. Its 2022 apartheid report cited language from the International Court of Justice's 1971 Namibia opinion as though the Court had supplied an international-law definition of apartheid based on race, color, descent or national or ethnic origin. But that passage wasn't defining apartheid at all. The Court was discussing South Africa's acknowledged racial policies in Namibia and whether they violated the UN Charter. Amnesty took language describing unlawful discrimination in one case and repurposed it as a definition of the crime it wanted to apply to Israel. I documented Amnesty's new “apartheid” definition before its report was released.
Then came genocide, and by that point the pattern was almost impossible to miss. Academics has already set the stage for the accusation before October 7. Activists promoted it immediately afterwards. Amnesty acknowledged the ICJ rule that, when genocidal intent is inferred from a pattern of conduct, genocidal intent must be the only reasonable inference. Amnesty then complained that this could be read “overly” narrowly and might make genocide difficult to find during armed conflict. So it adopted what it called a more “holistic” approach instead. Even more remarkably, Amnesty's holistic analysis treated Israel's supposedly unlawful occupation and supposed apartheid as part of the context establishing genocide—meaning its new genocide theory was partly resting on the earlier legal definitions it had already stretched to make Israel fit. As I wrote when the genocide report appeared, Amnesty had built one altered definition on top of the others.
That is not a minor technical dispute over terminology. It is an epistemic mechanism.
Start with a conclusion. Alter or expand “occupation” until Gaza remains occupied even after Israel's soldiers leave. Expand “racial group” and apartheid until Israel fits apartheid. Then use that apartheid finding as contextual evidence for genocide while loosening the standard for inferring genocidal intent. Each newly manufactured conclusion becomes a premise for the next one.
The house gets taller while nobody goes back downstairs to see whether the foundation was sound.
The same thing happens outside legal definitions. Amnesty's Gaza genocide report selected attacks in which it said it could find no military objective, and treated that absence as evidence supporting intentional attacks on civilians. But “Amnesty could not identify the target” and “there was no military target” are entirely different propositions. Subsequent examination of Palestinian sources has identified Hamas and Palestinian Islamic Jihad members among the dead in many of precisely those cases. The supposed evidence of inexplicable attacks was partly evidence of Amnesty's lack of information.
The famine narrative followed a similar pattern. Predictions of mass starvation generated enormous headlines; when the predicted death toll did not materialize, the underlying narrative barely changed. Disputed IPC methodology, missing denominators and subsequent nutrition data received nowhere near the propagation of the original accusation.
The epistemic environment does not merely amplify evidence. It determines which evidence remains visible.
The point is not that every accusation against Israel is false. Israeli soldiers, like all soldiers, do sometimes commit crimes. A tiny percentage of Israeli residents in Judea and Samaria commit crimes. Israeli governments sometimes make terrible decisions. The point is that the sweeping propositions used to define Israel—genocide, apartheid, deliberate starvation, systematic civilian targeting, ethnic cleansing—must be proved. They have not been; on the contrary, they have been proven false.
Instead, weak or fictional evidence supporting them is promoted, contrary evidence is marginalized, and sometimes the definitions themselves move.
Epistemic closure
Once enough weak claims acquire institutional approval, they become extremely difficult to dislodge.
For years, much of this process occurred inside academia, NGOs, fringe media and UN human-rights institutions. Those worlds increasingly cited one another, shared conceptual vocabulary and treated propositions established inside the network as foundations for further work. An academic could cite an NGO, the NGO could cite another academic, a UN expert could cite both, and the next academic could cite the UN expert. None of them noted that the original source is Hamas.
You now appear to have four independent confirmations. You may actually have one false assertion traveling in a circle.
At every step, the claim acquires another layer of institutional authority without necessarily acquiring another layer of evidence. An activist allegation becomes an NGO finding. The NGO finding becomes a UN expert's conclusion. The UN conclusion becomes something “the UN says.” A newspaper reports it as established background. Scholars then cite the newspaper and UN report as evidence of an emerging consensus.
The accusations reinforce one another as well. If Israel is already an apartheid state engaged in ethnic cleansing (let's not mention that one in five Israelis are Arab), genocide doesn't sound nearly as extraordinary. If Israel deliberately created a famine, that becomes evidence of genocidal intent. If Israel is already committing genocide, an Israeli strike that kills civilians is more readily interpreted as deliberately targeting them. Those supposedly deliberate attacks then become further evidence of genocide.
A conclusion built partly from bad evidence becomes a premise for interpreting the next piece of evidence.
“Settler violence” provides a smaller example. Crimes by some Israeli settlers exist and their perpetrators should be prosecuted. But the category itself is often analytically corrupt. Violence by at most a few hundred offenders is attached rhetorically to a population of roughly 700,000 Jews, while Palestinian violence is much more likely to be individualized if reported at all. Add selective reporting, disputed incidents and the absence of a denominator, and “settler violence” becomes evidence about “the settlers.”
This is why disproving individual claims becomes increasingly ineffective. I can show that famine predictions failed, that many purported civilians were Hamas fighters, that NGOs alter the legal standard they used, that one article omitted the denominator or that one supposedly independent source was quoting another source. But another twenty claims remain, and those claims protect the one just disproved. The reaction becomes: perhaps this particular story was wrong, but surely the overall picture is true. There is simply too much “evidence.” Where there's smoke, there must be fire. .
Except accusations are not the same thing as evidence.
Artificial intelligence gets sucked into precisely the same system. Ask an LLM whether Israel practices apartheid and it encounters Amnesty saying so, Human Rights Watch saying so, UN experts saying so, academics saying so, newspapers reporting that experts say so, and thousands of articles describing the accusation as increasingly accepted. The machine is very good at synthesizing its information environment. It is much less naturally equipped to recognize that ten prestigious sources may ultimately rest on the same two contested or refuted reports.
Worse, rebuttals don't propagate through the corpus nearly as efficiently. An Amnesty report is cited by thousands of scholars, NGOs and news outlets. A detailed rebuttal showing that Amnesty misread a legal text or failed to identify militants among the dead may be read by a few thousand people and disappear.
Corpus prevalence itself becomes a proxy for credibility. Garbage doesn't stop being garbage because it has been cited 5,000 times.
The larger danger begins when epistemic closure doesn't remain epistemic. Institutions that produce purported knowledge influence institutions that exercise power. Activists and politicians point to expert consensus. National governments point to NGOs, academics, UN officials and mounting political pressure. International institutions point to expanding governmental consensus. Each action then becomes a new fact capable of being fed back into the original system.
This week Britain provided an almost perfect example.
Britain: bad input becomes authoritative output
British Foreign Secretary Ed Miliband announced sweeping measures against Israeli settlements this week while saying that “the British government agrees” that ethnic cleansing is being perpetrated in parts of the West Bank by violent settlers. His statement also relied heavily on the familiar package of claims about blocked humanitarian aid, medical supplies, destroyed hospitals and Israeli violations of international law.
That is an enormous factual and legal conclusion built on extremely weak reasoning.
The existence of criminals within a population of hundreds of thousands proves the existence of criminals. To get from there to ethnic cleansing requires evidence of systematic removal, scale, intent, causation and responsibility. The phrase “settler violence” does much of that work rhetorically before the larger proposition has actually been proved.
The medical-aid claim illustrates the same problem. There can be shortages of particular medicines in Gaza. Distribution can fail. Hospitals can lack supplies. Aid organizations may be unable to collect what has crossed the border. None of those propositions means that Israel prohibits medicine from entering Gaza: it doesn't. Yet versions of “Israel blocks medicine” have circulated so widely that the distinction almost disappears.
That is the epistemic environment as an input. Miliband doesn't have to establish the propositions from scratch. “Settler violence,” ethnic cleansing, deliberate starvation, blocked medicine, destroyed hospitals and genocide arrive as an already assembled package, with each accusation lending plausibility to the others.
But Miliband is not an NGO researcher or activist. He is the British Foreign Secretary. Once he repeats these claims and Britain acts upon them, the claims change status.
The next NGO can cite British policy. The next journalist can describe Israel's growing diplomatic isolation. The next academic can write about expanding governmental recognition of Israeli abuses. The next activist can tell a city council that major Western democracies have already recognized what Israel is doing.
Bad evidence helped produce the policy. The policy then becomes better-looking evidence for the next claim.
There is also an obvious political incentive operating in Britain. Labour is competing for progressive voters with a Green Party and pro-Palestinian political forces demanding still more aggressive action against Israel. Once the epistemic environment provides politically useful accusations, the incentive is to escalate them rather than spend political capital examining whether they survive scrutiny.
The epistemic environment supplies the accusation. Domestic politics rewards using it. Government action then validates the accusation.
And once a politician adopts the strongest characterization, correcting it becomes politically expensive. If Miliband has declared ethnic cleansing, reconsidering whether the evidence actually establishes ethnic cleansing can be portrayed not as factual correction but as retreating from victims of ethnic cleansing.
The system acquires a ratchet.
Providence: designing the next input
Providence, Rhode Island, shows the same mechanism at a different level. An anti-Israel coalition has placed a divestment initiative on the November ballot. Its language does not mention Israel. It asks whether Providence should restrict investments involving foreign countries conducting “ethnic cleansing, apartheid, illegal military occupation, or genocide.”
Who could possibly object? I don't want my city investing in genocide either.
Then read the actual Providence divestment ordinance.
Who determines which countries qualify? The ordinance directs city officials toward UN Security Council resolutions; findings from UN Independent International Commissions of Inquiry; findings from any UN Special Rapporteur; findings from the UN High Commissioner for Human Rights; any U.S. court ruling; “definitions in U.S. Federal policy”; and International Court of Justice rulings and advisory opinions. Officials can also use “other reputable sources” in identifying complicit entities.
The Providence voter therefore isn't being asked to determine whether genocide, apartheid or ethnic cleansing has actually been proved. The entire epistemic question is outsourced.
Of course you should vote against genocide. But what voter will look at how that conclusion is determined?
According to an organizer quoted by the Boston Globe, the campaign wasn't aware of any country other than Israel that currently would qualify.
That should immediately raise questions. Myanmar has faced findings of genocidal intent against the Rohingya. The United States formally determined that the RSF and allied militias committed genocide in Sudan. The UN human-rights office concluded that abuses in Xinjiang may constitute crimes against humanity. Russia occupies Ukrainian territory after invading the country. Turkey has occupied northern Cyprus for decades.
Yet run those cases through Providence's exact trigger language and suddenly everything becomes technical. Does the finding come from exactly the right body? Does a U.S. genocide determination count as a “definition in U.S. Federal policy”? Does an authority have to use the exact words “illegal military occupation”? Do crimes against humanity count when the ordinance specifies other categories? What happens when approved authorities disagree?
Those ambiguities expose what is really unusual about the proposal. It does not ask the obvious universal question:
Which countries are committing grave human-rights abuses?
It effectively asks:
How can we define grave human-rights abuses through a particular collection of institutional labels so that the practical answer is only Israel?
And some of those authorities belong to the same epistemic environment whose work should be under examination. Providence explicitly authorizes reliance on findings from “any UN Special Rapporteur.” That includes Francesca Albanese, whose statements about the “Jewish lobby,” October 7 and Israel have themselves drawn accusations of antisemitism and bias from democratic governments.
Yet under Providence's system, her conclusions can potentially become municipal investment evidence.
This is not neutral outsourcing of expertise. It is outsourcing to a curated epistemic system whose assumptions and conclusions are then treated as the solution to the question they helped create.
Providence is not the endpoint
The amount of money Providence might divest is almost irrelevant. The important thing is whether the mechanism works.
If the initiative succeeds, activists have a template they can take to the next municipality: universal moral language for voters, selected institutional authorities in the fine print, and Israel as the intended practical target. If it fails, they learn which objections worked, modify the language and try somewhere else. (It seems likely that the same people pushing this initiative are the ones who, when in college, pushed student councils to pass BDS resolutions. The experience on campus is practice for local, and eventually national, action.)
Get enough cities to adopt variations of it and something new happens.
Look how many American cities are divesting from companies complicit in Israeli apartheid and genocide. Can they all be wrong?
Of course they can. Fifty municipalities do not represent fifty independent investigations of genocide. Municipal investment officials are not international-law specialists. They will rely on the NGO reports, UN findings, court decisions and “reputable sources” presented to them by advocates.
But fifty cities look like fifty data points.
The output has become another input.
The curve is accelerating
Britain and Providence are not two unrelated stories. They show two parts of the same feedback loop operating during the same week.
For much of the past twenty-five years, the anti-Israel epistemic environment primarily manufactured propositions: apartheid, settler colonialism, ethnic cleansing, deliberate starvation, illegal occupation, genocide. Many of those propositions rested on evidence that was weak, circular, selective, definitionally manipulated or simply wrong. But the institutional network allowed them to survive criticism and acquire prestige.
Now those propositions are increasingly manufacturing consequences: sanctions, trade restrictions, divestment measures, legal exposure and diplomatic isolation.
Consequences are much more powerful inputs than accusations. “Amnesty says Israel practices apartheid” is one thing. “Countries and cities around the world are sanctioning Israel because of apartheid” sounds entirely different.
But the second is simply a downstream consequence of the first.
It is no longer a simple chain. NGOs, academia, journalists, UN mechanisms, courts, local politicians, national governments and international institutions are increasingly capable of supplying apparent confirmation to one another from multiple directions at once.
That means the curve can accelerate.
It also means the system becomes progressively harder to challenge. Twenty years ago an academic could dispute whether apartheid was an appropriate analogy for Israel. Today he isn't simply arguing about the evidence. He is supposedly arguing against Amnesty, Human Rights Watch, UN officials, academic literature and years of newspaper coverage. Tomorrow he may be arguing against governments, courts, universities, pension funds and dozens of municipalities that have already acted on the proposition.
The same thing is happening with genocide.
“Israel is committing genocide” takes four words.
Explaining why that conclusion is unsupported requires discussing specific intent, alternative hypotheses, military objectives, casualty methodology, combatant classification, humanitarian aid, famine data, changed legal definitions, contradictory evidence and the provenance of institutional reports.
And none of that may matter rhetorically because the response takes another four words:
“You are defending genocide.”
That is one of the most dangerous products of the epistemic environment. Bad evidence no longer merely leads people to a false conclusion. Once enough institutions have endorsed that conclusion, questioning the underlying evidence itself becomes morally suspect.
Politicians, journalists and academics no longer have to be convinced that every accusation is true. They merely have to recognize that repeating it is safer than questioning it.
Back to Durban
This is why Durban looks more important twenty-five years later than it did at the time.
The NGO Forum wasn't describing an established international consensus. It was demanding one. It already had much of today's vocabulary: apartheid, ethnic cleansing, genocide and colonialism. It already had the desired policy response: boycotts, sanctions, embargoes and total isolation.
The evidence now used to justify those accusations came later, and a great deal of it has failed badly under scrutiny. It didn't matter. The goalposts were moved and the quality of "evidence" was trumped by the quantity.
The conclusions help determine the definitions. The definitions generate new conclusions. Those conclusions become context for still more accusations. It is circular construction.
And that is why, twenty-five years after Durban, we may still be looking at the start of the process rather than its culmination.
Providence and Britain are not results. They are new inputs.
The system is beginning to convert accusations into policy, and policy returns to the system as apparent evidence that the accusations were true. Every new adoption lowers the threshold for the next. Every policy creates another citation. Every citation strengthens the appearance of consensus. And every increase in consensus raises the social and political cost of going back to the beginning and asking the question that should have been asked before any of this happened:
What does the evidence actually prove?
That is the question that the accusers do not want anyone to answer. They consistently construct the arguments to begin with the accusations as true and build from there. Far, far more academic papers assume "genocide" than argue it. When pressed, most accusers point to "experts" and the "experts" point to their own expertise instead of evidence.
Only rarely do we see the questions: Did the event happen as described? Are the numbers meaningful? What is the denominator? Are combatants being counted as civilians? Does the source actually know what Israel was targeting? Is the organization applying the ordinary legal definition or constructing a special one? Is the supposedly independent source merely repeating another source? Was contrary evidence considered fairly?
The problem is not merely that the scales are weighted against Israel. The room has been built on a slope.
And the slope is getting steeper.
Thursday, September 10, 2026
Buy 
