Skip to content
Get in touchFR
Updated in October 2026
Explained

Should you boycott AI?
Would a boycott make any difference?

Calls to boycott AI, from an op-ed with 150 signatories to a teachers' manifesto and the QuitGPT campaign, cite 5 risks. We look at what can be measured for each, and at what a year of use adds up to for the climate.

Guillaume Pakula
By Guillaume Pakula, co-founder of Celsius. Since 2020, he has helped 80+ organisations with their Bilan Carbone® and climate strategy.
October 2026
Updated October 2026 · 12 min
Since late 2025, writers, researchers and activists have called for generative AI to be boycotted or refused. Their reasons fall into 5 risk categories. A climate consultancy can put a figure on one of them: on the climate, a year of regular use comes to 60% of a beef meal. The other 4 need to be assessed on criteria other than carbon.
Key takeaways
  • 1Calls to boycott or refuse AI cite 5 risks, only one of which can be measured in CO2e.
  • 2A year of regular AI use adds up to about 3 kgCO2e, or 60% of a beef meal (our calculation).
  • 3With 1 generated video per day, the year rises to 16 kgCO2e, more than 3 beef meals.
  • 4In 2025, 48% of people in France used generative AI, and search engines have since been showing it by default.

Since late 2025, several calls have urged people to boycott or refuse AI: an op-ed with 150 signatories in the French daily Le Monde; a manifesto by teachers and researchers, which has gathered more than 3,400 signatures; and a campaign to cancel ChatGPT subscriptions. These calls to boycott generative AI cite 5 risk categories: cultural, informational, societal, environmental and civilisational, of which only one can be measured in CO2e.

Boycotting AI

The 5 risks cited in calls to boycott AI

Choose an image: what the op-ed says, a dated fact and what can or cannot be measured. The order does not indicate their severity.

Creation and copyright

What the op-ed of 18 June 2026 says

It thrives on the massive theft of data, articles and works. (our translation)

A dated fact: 20 July 2026

A federal court in California approves the settlement ending the authors' lawsuit against Anthropic: $1.5 billion for 482,460 works, or about $3,000 per work.

Federal court for the Northern District of California, Bartz v. Anthropic

What can be measured

Amounts and numbers of works, when a settlement or a court sets them.

What cannot be measured

What an imitated style is worth; the works that will not be written.

Op-ed of 18 June 2026 (text published by its initiator; 3 of the 5 quotations cross-checked with AFP); federal court for the Northern District of California; Damien Charlotin; Stanford Digital Economy Lab; International Energy Agency (IEA); Center for AI Safety. Checked on 3 October 2026. "What can be measured" and "What cannot be measured": Projet Celsius's interpretation. Photos, from left to right: Nick-D (CC BY-SA 4.0), Martin Abegglen (CC BY-SA 2.0), Jérémy Barande / École polytechnique (CC BY-SA 2.0), Tony Webster (CC BY 2.0), UK Government (CC BY 2.0). Wikimedia Commons, cropped.

On the climate, a year of regular AI use comes to about 3 kgCO2e (kilograms of CO2 equivalent): 60% of a beef meal, according to our calculation. Giving up AI therefore does very little to reduce a person's carbon footprint; on a global scale, what matters is the growth in data centres. The other 4 risks cannot be counted in CO2e, but the calls cite them just as much.

1Calls to boycott AI: an op-ed, a manifesto and a cancellation campaign

An op-ed with 150 signatories, in June 2026

On 18 June 2026, Le Monde published an op-ed signed by 150 people, according to AFP (Agence France-Presse), including the writers Annie Ernaux and Hervé Le Tellier, the comic-book author Enki Bilal, the former French culture minister Françoise Nyssen and the biologist Marc-André Selosse.

The op-ed was initiated by Alexandre Basquin, senator for the Nord département, according to his own website, where he publishes the text. In our translation, it asks everyone who is not obliged to use generative AI to boycott it. Among the proposals that, in its words, exist, the text also lists rationing the energy resources allocated to generative AI.

A manifesto by teachers and researchers, late 2025

In late 2025, the Toulouse political ecology workshop (Atelier d'écologie politique de Toulouse, Atécopol) published a manifesto titled Face à l'IA générative, l'objection de conscience (Faced with generative AI, conscientious objection, in our translation). Its signatories, mostly from higher education, research and the French national education system, commit not to use these tools in their teaching, research or administrative work, unless expressly obliged to do so. According to La Relève et la Peste, a French news site, the manifesto had more than 1,300 signatories in early December 2025; according to the French news site Mediapart, it had more than 3,400 signatures in early July 2026. The text speaks of conscientious objection; the word "boycott" comes from press coverage.

Boycotting AI

From the 2023 call for a pause to the 2026 call for a boycott

8 milestones on a time scale: warnings, collective actions, then calls to boycott or to refuse generative AI.

18 June 2026 · Boycott or refusal

150 signatories call for a boycott in Le Monde

Writers, artists, scientists, former ministers

  • Boycott or refusal
  • Warning
  • Collective or legal action
Future of Life Institute; Center for AI Safety; Writers Guild of America; SNE, SGDL and SNAC (French publishers' and authors' bodies); Atécopol; Indie Game Awards; QuitGPT; AFP. Texts and press reports collected on 3 October 2026.

A cancellation campaign and a video game awards decision

In the United States, the QuitGPT campaign, organised in late January 2026, calls on people to cancel their ChatGPT subscriptions. Its organisers first criticise OpenAI for a donation by its president, Greg Brockman, to the political committee MAGA Inc., according to MIT Technology Review. They then criticise it for the agreement announced on 28 February 2026 with the US defence department, which they say could lead to mass surveillance and autonomous weapons; OpenAI states that the agreement rules out these 2 uses.

According to Euronews, the campaign claimed more than 1.5 million participants on 2 March 2026. The figure adds together cancellations, social media shares and sign-ups on its website.

Refusal also takes the form of rules. On 20 December 2025, the Indie Game Awards withdrew 2 awards from the French game Clair Obscur: Expedition 33, because the awards' rules exclude games developed with generative AI. According to Next, a French technology news site, the studio explained that the issue concerned placeholder textures, which were removed and replaced by the final textures.

The calls do not all say the same thing. We sort their reasons into 5 risk categories, which the op-ed brings together in a single text. For each category, we give what the op-ed says (quoted from the text published by its initiator, in our translation), a fact, and what can and cannot be measured. The order is that of our framework and says nothing about how serious each risk is.

Cultural: creation and copyright

The op-ed says that generative AI thrives on the mass theft of data, articles and works. On 20 July 2026, a federal court in California approved the settlement ending the lawsuit brought by authors against Anthropic over the unauthorised use of their works to train its models, according to the Authors Guild, an organisation of authors. The settlement amounts to $1.5 billion for 482,460 works, or about $3,000 per work, according to the order.

Shelves of old books running the full height of a room, with a ladder and a wrought-iron gallery
The shelves of the Mazarine Library in Paris, in 2025. The settlement approved in July 2026 between Anthropic and authors covers 482,460 works, for $1.5 billion. Photo by Rémih, Wikimedia Commons, CC BY-SA 4.0, cropped

Informational: errors, fakes and trust

The op-ed denounces a mush of synthetic images and deepfakes. In June 2023, a New York court sanctioned lawyers who had cited 6 court decisions that ChatGPT had invented, and Damien Charlotin's database listed, as of 12 November 2026, 2,292 decisions in which a court noted content invented by an AI. As for fakes, CNN reports that in January 2024 an employee of the engineering group Arup in Hong Kong transferred the equivalent of $25.6 million after a faked video conference.

Societal: employment and learning

The op-ed predicts millions of jobs destroyed and, among young people, weakened cognitive abilities. In US payroll records, employment among workers aged 22 to 25 in the occupations most exposed to AI is 19% lower than it would have been had it tracked employment in other occupations, according to a Stanford study (August 2026); its authors do not establish the cause.

On learning, an MIT study of 54 participants, which has not yet been peer-reviewed, observes lower brain connectivity among those who write with an assistant; the study team asks people not to read this as meaning that AI damages the brain. See AI and jobs and AI and the brain.

Environmental: energy, water, metals

The op-ed writes that the frenzy around generative AI marks a clear renunciation of the great battle of our century, ecology. It is the only one of the 5 risks that is counted in kilowatt-hours and CO2e. See all the environmental impacts of AI and the water AI consumes.

Civilisational: loss of control

On 30 May 2023, a 22-word statement from the Center for AI Safety put the risk of extinction from AI alongside pandemics and nuclear war. The statement had more than 350 signatories, according to the New York Times, including the heads of OpenAI, Google DeepMind and Anthropic. In April 2025, Geoffrey Hinton, winner of the 2024 Nobel Prize in Physics, put the risk that AI would end up taking control from humans at between 10 and 20%, according to CBS News.

Geoffrey Hinton standing at a lectern, hand raised, during his Nobel lecture in Stockholm
Geoffrey Hinton during his Nobel lecture in Stockholm on 8 December 2024. He had left Google in May 2023 so that he could speak about the dangers of AI without implicating his employer. Photo by Jay Dixit, Wikimedia Commons, CC BY-SA 4.0, cropped

The risk of loss of control is disputed. In October 2024, Yann LeCun, Turing Award winner, then a researcher at Meta and a professor at New York University, called these fears absurd ("complete B.S.", in his words), according to TechCrunch. As early as March 2023, in response to a letter calling for a pause, the DAIR research institute, founded by Timnit Gebru, wrote that these hypothetical risks distract from the real harms of systems already deployed.

The text of the op-ed does not mention a risk of extinction. It condemns a vision of society founded on the marginalisation of human beings and the destruction of our living environment.

3The environmental risk in orders of magnitude

A year of regular use: 60% of a beef meal

For 10 questions per day, plus 5 images and 1 video of 5 seconds per week, the year comes to 3 kgCO2e with the global average electricity mix: 60% of the emissions of a beef meal, or those of 21 km by car, according to our calculation based on reference values from ADEME (the French Agency for Ecological Transition). It would take 690 years of this use to equal a Paris-New York return flight; see what a flight represents in a carbon budget. For one person, giving up AI makes very little difference to their carbon footprint.

Boycotting AI

A year of your AI, on a scale that runs up to a Paris-New York round trip

Set your usage and see where the total falls, next to a meal, a smartphone, a car and a flight.

One year of your AI

3kgCO2e

That is 60% of the emissions of a beef meal, or 21 km by car.

0.04% of the annual carbon footprint of a person in France.

  • Questions500 g
  • Images350 g
  • Videos2.1 kg
Server electricity

The full scale, from 0 to 2,060 kgCO2e

New smartphone: 80 kg1,000 km by car: 142 kgParis-New York return flight: 2,060 kg
Your year of AI: 3 kg

Close-up: the first 10 kg, magnified 206 times

Beef meal: 5 kg
Your year of AI: 3 kg
Epoch AI (2025); Luccioni et al. (2024); Delavande et al. (2025); Ember, Global Electricity Review 2026 (AI Impact Database); ADEME (the French Agency for Ecological Transition), Impact CO2; SDES (the French Ministry of Ecological Transition's statistics service). Projet Celsius calculation, server electricity only: hardware manufacturing and model training are not counted.

This calculation counts only the electricity used by servers. Using the life cycle assessment figure (in French) that Mistral AI has published for one response (1.14 gCO2e, including server manufacturing), the emissions from the questions are more than 8 times higher and the year rises to 6.7 kgCO2e, or 1.3 beef meals: the same order of magnitude. See why AI is bad for the environment, from the chip to the response, and how much energy a question put to ChatGPT uses.

Video changes the calculation

Of this total, the 52 videos account for 2.1 kgCO2e, the 3,650 questions for 0.5 and the 260 images for 0.3. With 1 video per day, the year rises to 16 kgCO2e, or more than 3 beef meals, of which the videos account for 15. The generator used as the reference in this series of articles is one of the most energy-hungry of those measured; with a median generator, the year of regular use falls to 1.5 kgCO2e. See the footprint of an AI-generated video and that of an image.

Boycotting AI

Depending on the assumptions, a year of AI emits between 0.27 and 16 kgCO2e

Our regular-use profile, changing one assumption at a time. In kgCO2e per year.

1 beef meal
Epoch AI (2025); Luccioni et al. (2024); Delavande et al. (2025); Ember, Global Electricity Review 2026 (AI Impact Database); ADEME, Impact CO2; Mistral AI (2025). Projet Celsius calculation.

At system level, consumption that could almost double by 2030

The environmental argument made in the calls is aimed at the system. The world's data centres consumed 485 TWh of electricity in 2025 (terawatt-hours, that is billions of kWh), and the International Energy Agency (IEA) expects 945 TWh in 2030. At this scale, what matters is the number of centres built, the electricity that powers them and the uses built in by default, far more than any individual user's choice. See how much electricity AI uses.

4Can a boycott of AI make a difference when AI is widely used and built in by default?

Nearly half of people in France already use generative AI

According to the Baromètre du numérique (the French digital barometer), 48% of French people aged 12 and over used generative AI (in French) in 2025, compared with 20% in 2023, and the share reached 85% among 18 to 24 year-olds. In the same survey, 46% consider that it has a greater environmental impact than a conventional search.

An OpinionWay poll commissioned and presented in May 2026 by CeSIA (Centre pour la sécurité de l'IA, a French organisation), and cited by the op-ed, found that 8% of French people want to speed up the development of AI in France and Europe, while 42% want to slow it down sharply or pause it temporarily.

About ten protesters hold signs calling for AI to be regulated, beside the water in Berlin
A rally by the PauseAI movement in Berlin on 8 February 2025, to coincide with the AI Action Summit in Paris. In France, 42% of respondents want to slow the development of AI sharply or pause it temporarily, according to an OpinionWay poll presented in May 2026. Photo C.Suthorn / cc-by-sa-4.0 / commons.wikimedia.org (CC BY-SA 4.0), cropped

AI built into search engines, messaging apps and software

You can do without an assistant; AI built in by default is harder to avoid. Google launched its AI Overviews in France on 22 July 2026: AI-generated answers appear in the results, and its blog post mentions no setting to turn them off.

Meta has been rolling out its assistant on WhatsApp in 41 European countries since March 2025; the Italian competition authority opened an investigation in July 2025 for suspected abuse of a dominant position, noting that the assistant had been pre-installed without users having asked for it. It closed its proceedings in May 2026 without ruling: the European Commission had extended to Italy its own investigation into a Meta rule closing WhatsApp to competing assistants. On 9 June 2026, the Commission ordered Meta to reopen that access provisionally, pending its final decision.

Microsoft included its Copilot assistant in its consumer subscriptions in January 2025, with a price increase of €30 a year in France, according to Next. According to Microsoft, a setting lets you turn it off, and existing subscribers could switch to a plan without Copilot.

What a boycott can achieve, and what it does not reach

An individual boycott sends a signal, especially when it targets a paid subscription: that is QuitGPT's bet. It does not reach AI installed by default or corporate uses. Clubic, a French technology site, notes that the general public is only one part of the market (our translation). On the climate, its direct effect is small; on the other 4 risks, it expresses a refusal that emissions figures do not measure. None of this makes a boycott useless: its effect is felt elsewhere than in a person's carbon footprint.

5How to govern the use of AI

Between using nothing and using everything, there are 3 levels of action: individual choice, refusal within a profession and collective rules.

Boycotting AI

From individual choice to collective rule: 3 levels of action

What to expect from each level, and its limit.

Level 1Individual choice

A signal sent to vendors, especially when it targets a paid subscription.

Examples: Cancelling a subscription (QuitGPT); following the op-ed's call to users; being selective about one's uses

Level 2Refusal within one's profession

A professional standard, visible to the public: a label, a statement, award rules.

Examples: The manifesto of teachers and researchers; the rules of a video game awards

Level 3Collective rules

Enforceable rights: not having the tool imposed, being informed, being able to object or to switch it off.

Examples: The 2023 agreement of American screenwriters; the opt-out form obtained from Meta in Europe; what the op-ed asks of legislators
Examples: QuitGPT, 18 June 2026 op-ed, Atécopol, Indie Game Awards, Writers Guild of America, Irish data protection authority. Projet Celsius interpretation.

Individual choice: boycott AI, or use it selectively

At this level, you can either boycott AI or use it selectively. In our ranking of 10 ways to reduce the impact of AI, the top 3 actions are doing without generated video, asking yourself whether AI is needed and keeping reasoning mode (where the model works through steps before answering) for difficult questions. The choice of service also matters: see which eco-friendly AI chatbot to choose.

Refusal within one's profession

The Atécopol manifesto is a professional refusal: its signatories commit to it in their teaching and research, and say so publicly. The Indie Game Awards wrote it into the rules of their awards. In France, the publisher Éditions Double Ponctuation announced in November 2024 that it was putting on its books a logo reading, in our translation, guaranteed free of artificial intelligence, according to ActuaLitté, a French publishing news site. These refusals create a norm visible to the public in settings where no employer or client imposes the tool.

Collective rules

In 2023, after 148 days on strike, American screenwriters secured an agreement under which a studio cannot impose AI software on them and must tell them about any AI-generated material, according to the summary of the agreement; the next contract, in force until 1 May 2030, keeps these protections according to the Writers Guild of America. In Europe, the Irish data protection authority obtained from Meta in May 2025 a simplified objection form, ahead of the training of Meta's models on users' public posts.

A march of striking screenwriters behind a black Writers Guild banner, red signs held up
Writers Guild of America screenwriters on strike on 21 June 2023; one sign refers to AI. The agreement signed after 148 days of strike action prohibited a studio from imposing AI software on a writer; the next contract, in force until 1 May 2030, keeps this rule according to the union. Photo UFCW 770, Wikimedia Commons, CC BY 2.0

In France, ADEME has asked since July 2026 for users to be able to switch off, if they wish (our translation), the generative AI functions added to existing services. For a company, the rule starts with a usage policy and a measurement: our AI footprint calculator gives an order of magnitude, and our guides explain how to count AI in your Bilan Carbone® (the French carbon accounting method) and measure the AI footprint of a company.

6Key takeaways

  • Calls to boycott or refuse AI cite 5 risks: cultural, informational, societal, environmental and civilisational.
  • A year of regular use comes to about 3 kgCO2e, 60% of a beef meal; video accounts for most of it.
  • The other 4 risks cannot be measured in CO2e: they concern copyright, information, jobs and loss of control.
  • Nearly half of French people already use generative AI, and it is arriving by default in search engines and messaging apps.

Further reading: digital pollution statistics.

Further resources

Frequently asked questions

or: guillaume@projetcelsius.com

More articles to read

View all →
French environmental cost label for clothing: what changed on 1 October 2026
Regulation

French environmental cost label for clothing: what changed on 1 October 2026

6 min read
What is the French textile eco-score?
Who displays it, and has it become mandatory?
Regulation

What is the French textile eco-score? Who displays it, and has it become mandatory?

12 min read
Removable batteries and battery passports: who is affected in 2027?
Regulation

Removable batteries and battery passports: who is affected in 2027?

14 min read

A senior consultant reads every message

Tell us where you stand and what is pressing, and we will reply within 24 working hours with a first view on how to proceed.

Write to a consultant →
Our toolkit

A full range of free assessment tools

Cost estimator, obligation and eligibility checkers, footprint calculator.

Estimate costs and impacts
Ref. 2026·BC·0184REPORT · FINANCIAL YEAR 2026Bilan Carbone®complete 1·2·3Presentation to the Executive Committee · 2027 action planCONTENTS01Executive summaryp. 402Scope and methodologyp. 1203Emissions by scopep. 24042027 action planp. 5605Methodological appendicesp. 78VOLUME84 pages · ConfidentialSTANDARDISO 14064 · GHG ProtocolQUOTATION · BC.2026.0184Issued 05·05·2026 · Valid for 30 daysCelsius consultancy · Paris 3rd arrondissementINDICATIVE COST ESTIMATEYour companyMid-sized company · 180 employees · CSRD 2027SERVICEBilan Carbone®Scope 1·2·3 · 13 weeksBPIFRANCE FUNDING · - 60%INDICATIVE RANGE EXCL. VAT€16,800to €22,400PROJECT DETAILS01 · SCOPING3 weeks~€4,20002 · MODELLING8 weeks~€9,80003 · PRESENTATION2 weeks~€4,800Indicative estimate · Refined after scopingSIRET 891 234 567 00012

Bilan Carbone® cost estimator

Your price range in 1 minute, based on 2026 market rates and public funding.

1 minStart
Estimate costs and impacts
AI FOOTPRINT · MONTHLY USEYOUR TOTAL FOOTPRINT4.2 kgCO₂eq for 1,000 typical prompts · May 2026MODEL COMPARISON · 1,000 TYPICAL PROMPTSGPT-4oOpenAI6.8 kgClaude OpusAnthropic2.1 kgGemini ProGoogle4.5 kgMistral LargeMistral AI · FR1.6 kg

AI carbon footprint calculator

The climate impact of your AI queries, by model and task. Ecologits methodology.

3 minStart
Check an obligation
DIAGDÉCARBON'ACTIONACCELERATE THETRANSFORMATIONOF YOUR COMPANYIN ENERGY ANDTHE ENVIRONMENTOFFICIAL SCHEME · BPIFRANCE × ADEMECHECKER · ELIGIBILITYELIGIBLEYou meet the 3 Diag Décarbon'Action criteria.YOUR FUNDING€10,000 excl. VAT€6,000excl. VATAmount payable by you after the Bpifrance subsidyFUNDING BREAKDOWNBPIFRANCE 40%YOU 60%€4,000€6,000SCHEME SUBSIDISED BY

Diag Décarbon'Action eligibility

Check in 30 seconds whether your business is eligible for Bpifrance funding covering 40% of a Bilan Carbone® assessment.

30 secStart
Check an obligation
FRENCH REPUBLICMinistry for the Ecological TransitionLEGAL REQUIREMENTGreenhouse GasEmissions AssessmentArticle L. 229-25 of the French Environmental CodeDecree 2022-982 · published 1 July 2022FREQUENCY · 4 YEARSFILING · ADEMECHECKER · BEGES REQUIREMENTREQUIRED720 employees · BEGES requirementPOTENTIAL PENALTY€50,000maximum fine for failing to complete itart. L. 229-25OFFICIAL REFERENCE

BEGES checker

Does the BEGES requirement apply to you? An immediate answer, with the deadline and penalty.

30 secStart
Check an obligation
9:42SCANSCANNING DPP...DPP IDENTIFIEDOrganic cotton T-shirtSKU TX-CB-220 · Batch L-26-04781ESPR COMPLIANT42 DATA POINTS · 5 SECTIONSORIGINIndia · GOTSOrganic cottonFOOTPRINT5.2 kg CO₂eqLCA ISO 14040RECYCLABILITY85%Cotton recycling schemeREPAIRABILITY7.5 / 10Accessible partsEND OF LIFETextile recycling scheme (TLC) · Bordeaux/FRRefashion-approved recovery operatorEU · ESPR 2024/178112.05.2026

DPP checker

Is your product covered by the Digital Product Passport?

1 minStart
Check an obligation
COSTENVIRONMENTAL386POINTS257PER 100 GOfficial methodEcobalyse, v7.0.0CHECKER · ENVIRONMENTAL LABELLINGELIGIBLETextiles · voluntary labellingENVIRONMENTAL COST386 ptsT-shirt 150 g · 257 pts per 100 gExample: Ministry for the Ecological Transition FAQ

Textile environmental labelling 2026

Voluntary labelling, but third parties can publish it from October 2026: where do you stand?

1 minStart