Executive Summary

Millions of users canceled ChatGPT following reports of Pentagon cooperation, switched to Claude, and celebrated themselves as morally superior. Yet tech entrepreneur Malcolm Verrota deconstructs a fundamental misconception in his analysis: the entire Western AI infrastructure – Microsoft, Google, Anthropic, all frontier labs – has been working with the U.S. military for years. Canceling ChatGPT while daily using Copilot, Azure, Gmail, and GitHub is symbolic outrage without effect. The problem is not a single company, but an integrated system from which there is no escape.

People

Topics

  • AI-military cooperation
  • Hypocrisy in tech boycotts
  • Cloud infrastructure and data surveillance
  • EU data protection vs. reality
  • State surveillance technologies (Pegasus, trojans)

Clarus Lead

The #CancelChatGPT movement exposes itself as performative morality. While millions canceled their ChatGPT subscriptions, all frontier AI labs (OpenAI, Anthropic, Google, Anthropic) have been working with the Pentagon for years – for approximately $200 million per company. Users switching to Claude don't realize that Claude was first trained on classified Pentagon classification systems and today is actively deployed in operations against Iran. The actual risk lies not in a single application, but in the integrated ecosystem: Microsoft Copilot runs on the same GPT models as ChatGPT, Azure Cloud hosts Pentagon data, GitHub stores code under Microsoft control. Anyone truly rejecting this system would have to quit their job.

Detailed Summary

The Myth of the Ethical Alternative

The central fallacy: Claude is sold as a morally clean alternative. In reality, Claude was the first frontier AI model to run on classified Pentagon networks – for over a year, funded with $200 million. Anthropic was only recently dropped from the deal, not because the company had a principle, but because it refused to remove its safety measures against fully autonomous killer drones and mass surveillance of U.S. citizens. OpenAI was less hesitant: Sam Altman signaled willingness to both.

The Sprawling Pentagon Ecosystem

Microsoft invested $13 billion in OpenAI to run GPT models on Azure. In April 2025, the U.S. Department of Defense authorized Microsoft Copilot for the highest classification level (Impact Level 6): the CIA, NSA, and the Department of War use it routinely for targeting. The model is identical to ChatGPT. OpenAI's CEO, Paul Nakasone, was director of the NSA – the equivalent of a former BND chief sitting on SAP's board.

Anyone canceling Copilot would have to shut down Microsoft Windows, Office 365, OneDrive, Excel, PowerPoint, GitHub, and Azure. Impossible for 99% of employers. This is not malice, but system architecture.

The Swiss Privacy Illusion

The Swiss pride themselves on a data protection culture. Proton Mail from Geneva was considered a bulwark against U.S. surveillance – until the Swiss government revised the VUEPF ordinance in 2025. New rule: providers with over 5,000 users must store IP addresses for 6 months and verify user identities. Even worse: Article 50a requires encryption services to remove encryption on demand – something even Telegram or Meta don't do.

Andy Yen, CEO of Proton, compared Switzerland in 2025 to Russia in terms of data protection. His plan: relocate Proton infrastructure to Germany and Norway (€100 million cost). Government requests increased from 26 (2017) to 11,000 (2024) per year – an increase of 400 percent in seven years. Proton's compliance rate: 94%. OpenAI received 309 government requests in the same period (with 800 million weekly users).

The European Surveillance Escalation

Germany: The BND monitors over one trillion Internet transactions daily and can intercept 30% of global telecommunications. The Federal Constitutional Court declared BND surveillance unconstitutional in 2020 and 2024. The government's response: expanded powers.

Austria: In July 2025, the country passed a state trojan law allowing intelligence agencies to legally deploy spyware on phones and computers. The Constitutional Court had declared a similar law unconstitutional in 2019.

EU Pegasus Network: 14 EU states purchased Israeli spyware Pegasus. Documented victims: Jeff Bezos, Emmanuel Macron, opposition politicians in Spain. Greece approved over 15,000 Pegasus installations per year (2021).

The ETF Hypocrisy

Users who cancel ChatGPT but maintain Vanguard ETF savings plans (€500/month) unwittingly finance the same system. The Vanguard S&P 500 ETF contains Microsoft, Nvidia, Palantir (Pentagon AI), Lockheed Martin, RTX, and other defense companies. The Vanguard Information Technology ETF (VGT) consists of 32% semiconductor companies that train Pentagon AI. This is retirement savings with integrated weapons manufacturers.

Why Local Open-Source Models Don't Work

Three reasons: (1) 99% of the population cannot set up local LLMs – it requires 16-48 GB VRAM, Linux knowledge, Ollama, and LM Studio. (2) Frontier models like Claude Opus are significantly superior to frontier-closed-source models. Companies use them anyway for quality reasons. (3) The problem is not the model, but the ecosystem: emails are with Microsoft/Google, files in the cloud, code on GitHub, meetings in Teams/Zoom, CRM in Salesforce. All are intertwined with Pentagon infrastructure.

Key Points

  • All frontier AI labs work with the Pentagon: OpenAI, Anthropic (Claude), Google, XAI. Canceling ChatGPT and switching to Claude means switching to an earlier and more active Pentagon AI.

  • Microsoft Copilot is identical to ChatGPT, runs on Azure, and is used by the CIA/NSA for classification level 6 operations. Those who cancel Copilot lose their job; those who cancel ChatGPT continue using the identical model via Copilot.

  • Western state data surveillance exceeds Pentagon surveillance: Switzerland (VUEPF), Germany (BND: 1 trillion transactions/day), Austria (state trojans from 2025), 14 EU states (Pegasus spyware). European privacy is myth.

  • ETF savings plans fund defense AI: Vanguard S&P 500, MSWorld contain Microsoft, Nvidia, Palantir, Lockheed Martin. Moral consumers finance the system they boycott.

  • Real regulation instead of app-store activism: Colorado AI Act, Local Law 144 (New York), EU AI Act (August 2026) demand transparency and accountability – not from OpenAI, but from employers and governments.

Critical Questions

  1. Evidence: Verrota claims all frontier AI labs have $200 million contracts with the Pentagon. Are these contracts publicly viewable or only known through leaks? How reliable is the data basis?

  2. Conflicts of Interest: Verrota founded the "Chief AI Academy" (€1,500 course). Whose interests are served by the message "stop feeling guilty about AI use" – the user's or AI providers'/Verrota's own?

  3. Causality: The claim that "canceling ChatGPT while using Copilot is ineffective" assumes user boycotts are systemically irrelevant. Can isolated boycotts not at least create reputational risks for providers that force change?

  4. Alternatives: If Western AI infrastructure cannot be avoided, what practical actions does Verrota propose that are not "ineffective"? The EU AI Act is top-down; can citizens actively influence it?

  5. Implementation of Transparency: The EU AI Act requires transparency reports from employers. Who monitors compliance? What penalties exist for violations, and are they sufficiently deterrent?

  6. Switzerland-Specific: Proton's departure and VUEPF revision are real, but VUEPF is not yet law. How certain is the claim that Switzerland already monitors like North Korea – is this status quo or prognostication?

  7. Pegasus Scaling: 14 EU states bought Pegasus, but do all use it on Greece's scale (15,000/year)? Which countries use it most, and against whom?

  8. Counter-Hypothesis: Could Pentagon AI infrastructure – despite military use – be strategically indispensable for EU states to not fall behind China/Russia? Would a boycott actually cause more harm than good?

Reference List

Primary Source: Verrota, Malcolm (2026): "ChatGPT kündigen, Copilot nutzen – Das KI-Heuchel-Paradoxon." Podcast episode, Verrota AI Studio, Bregenz, March 6, 2026.

Verification Status: ✓ 2026-03-06 (based on podcast transcript)

Note: The episode contains numerous factual claims regarding Pentagon contracts, surveillance laws (VUEPF, BND, state trojans, Pegasus), ETF compositions, and Proton compliance data. Detailed source verification would require official government documents, transparency reports (OpenAI, Anthropic, Proton), and government decisions. This text summarizes Verrota's argument; individual facts should be verified independently.

This text was created with the support of an AI model. Editorial responsibility: clarus.news | Fact-check: 2026-03-06