Why Activists Must Harvest Big Tech’s Subsidized AI — Before They Fence It Off
I want to tell you about a window. It is open right now. It will not stay open. And most of the people who need to climb through it — the activists, the human rights defenders, the digital rights organizations, the nonprofits trying to save what is left of our shared humanity — are standing outside it, debating whether the window is ethical.
I understand the debate. I have spent thirty years inside this industry and the better part of a decade documenting its crimes. But I have also watched what happens when we hesitate while the other side does not. So let me make the argument plainly.
Remember When Search Was Plural?
In the late 1990s, we had choices. AltaVista. Lycos. Excite. HotBot. Ask Jeeves. Directory services like LookSmart and the Yahoo Directory, where human beings — actual human beings — organized the web by hand. If one engine buried your website, another would find it. If one directory rejected you, you knocked on the next door. The information ecosystem was imperfect, chaotic, and plural. Plurality was its protection.
That world is gone. As of mid-2026, Google holds roughly 90 percent of global search — StatCounter puts it at 91.25 percent of worldwide search referrals as of June 2026, with Bing a distant second under 5 percent. Two companies now decide, for nearly the entire connected planet, what is findable and what is functionally invisible. And this is not my opinion or my grievance. In August 2024, a federal judge ruled that Google had illegally monopolized the search market — the most significant antitrust ruling against a technology company since the Microsoft case of the late 1990s.
Think about what that consolidation cost us. Every story I have documented — the censorship of Palestinian journalists, the deletion of 700 videos of human rights evidence from YouTube, the erasure of entire human rights organizations from the platforms — was made possible because the doors narrowed to a hallway, and the hallway narrowed to a keyhole, and one or two companies now hold the key.
We Are Living Through the AltaVista Moment of AI
Now look at artificial intelligence in 2026. ChatGPT. Claude. Gemini. Grok. DeepSeek. Mistral. Llama. Perplexity. Many doors. Many engines. Competing companies burning billions of dollars of venture capital to win your loyalty, which means the tools are cheap or free — subsidized not by advertising yet, but by investors betting on future dominance.
This is the AltaVista moment. This is 1998. And if the history of search teaches us anything, it teaches us this: the plurality is temporary. The subsidy is temporary. The generosity is a customer-acquisition strategy, not a philosophy. A handful of dominant players will consolidate, the walled gardens will rise, the prices will follow, and the terms of service will be rewritten by the same lawyers who rewrote OpenAI’s terms in early 2024 — when the company quietly deleted its ban on military use of its technology, and the Israeli military walked through the open door via Microsoft’s Azure platform.
Yes, open-source models exist, and I support them. But let us be honest with each other, because honesty is the entire point of my work: without large-scale compute, open models cannot do what the frontier systems do. They cannot search the live web at scale. They cannot process the volumes of documents that investigations require. Openness without compute is a garden without water. The compute belongs to the same five companies I have spent two books naming.
They Are Already Using It. Against Us.
While nonprofits debate whether touching AI is pure enough, governments are pouring millions into bending it.
This is not speculation. It is in filings with the U.S. Department of Justice. In September 2025, Foreign Agents Registration Act documents confirmed that the Israeli government — through its Foreign Ministry and a $145 million public diplomacy initiative called Project 545 — hired a network of American firms for one of the most ambitious AI influence operations ever documented. At its center: Clock Tower X LLC, the firm of Brad Parscale, digital architect of the 2016 Trump campaign, under a $6 million contract routed through Havas Media Network on behalf of Israel’s Ministry of Foreign Affairs. The contract’s statement of work included a clause to deploy “websites and content to deliver GPT framing results on GPT conversations” — a minimum of 100 creative assets per month, 50 million paid impressions per month, optimized by predictive AI to dominate what both human beings and AI training systems encounter first.
Let me be precise, because I deal in documented facts and my critics are waiting for me not to: this was not a payment to OpenAI. Snopes investigated the viral rumor and found no contract between Israel and OpenAI. What the FARA filings document is something more sophisticated and harder to fight — a state-funded operation to poison the well that AI drinks from. To flood the training environment. To teach the machines what to believe, so the machines can teach you.
We saw this playbook on Wikipedia, where coordinated editing campaigns fought for years over how occupation and war would be described. Now the same battle has moved to the training data of systems that hundreds of millions of people treat as their primary source of truth. The propagandists did not hesitate. They did not hold a panel discussion about whether AI was problematic. They wrote a check.
The Objection I Hear — and Why It Fails
Every time I make this argument, someone raises the environmental cost of AI, the labor exploitation behind it, the harms of the industry that built it. I do not dismiss any of that. I wrote a chapter about the invisible workforce. I know what the data centers consume. These harms are real, and the people fighting them are right to fight.
But here is the fallacy hiding inside the objection: the harm of AI does not come from a human rights organization using a chatbot to draft a grant application. The harm comes from surveillance capitalism and the waging of war. It comes from targeting algorithms in Gaza. It comes from Pegasus on a journalist’s phone. It comes from a $1.2 billion Project Nimbus contract, from AI analyzing drone footage, from influence operations gaming the machines. Refusing to use a subsidized tool for thought leadership does not reduce a single one of those harms by a single gram. It only guarantees that the machines are trained on their narrative and never on ours — that the record we keep at ParentsPlea.com, the testimony in Forever Peace Now, the documentation in No Ethics in Big Tech, remains outside the well while their poison fills it.
An activist boycotting the drafting of a press release does not slow the war machine. It just makes the press release later.
Harvest the Bubble
So here is my call, and it is the same call I made in “Harvest the Bubble”: use it. Now. While it is cheap. While it is plural. While the venture capitalists of Silicon Valley are, for once in their lives, accidentally subsidizing the work of humanity.
Use it to write the grant applications that keep your organization alive. Use it to translate testimony into twenty languages. Use it to process the documents, structure the databases, draft the petitions, build the websites that no foundation would fund. Use it to produce the thought leadership that answers, point for point, the manufactured content that a $145 million operation is pumping into the training data. Every document we publish, every profile we record, every fact we anchor to a source is a drop of clean water into the well they are trying to poison.
The war profiteers looked at Gaza and saw what they called a golden opportunity. I have spent a chapter of my book on that phrase and what it cost in human lives. So let me take the phrase back from them: this window — this brief, subsidized, plural moment before the walled gardens rise — is our golden opportunity. Not to profit. To document. To testify. To organize. To make the record so large, so sourced, and so human that no algorithm can erase it and no influence operation can outweigh it.
In 1998 we did not know we were about to lose the plural web. We have no such excuse in 2026. The window is open.
Climb through it.
Vahid Razavi is the author of “No Ethics in Big Tech” and “Ethics in Tech and Lack Thereof”, and the director of the documentary “Forever Peace Now”. His work is self-funded — no corporate sponsors, no government grants, no advertising. Learn more at NoEthicsInBigTech.com, ForeverPeaceNow.com, ParentsPlea.com, and MyAWSStory.com.
Sources: StatCounter via Statista, global search engine referral share, June 2026; U.S. v. Google, ruling of August 2024; Responsible Statecraft, “Israel Wants to Train ChatGPT to Be More Pro-Israel,” October 2025; Sludge, “Israel’s New U.S. Influence Campaigns Target TikTok, Churches, and ChatGPT,” October 2025; Snopes, “Israel Didn’t Sign a Deal With ChatGPT. Here’s What Actually Happened,” December 2025; Associated Press investigation on Israeli military use of OpenAI models via Azure, February 2025.