Google AI workers in pioneering union bid over Israel and U.S. military tech use

98% of Google DeepMind members vote YES.

Scores of Google AI workers have launched an unprecedented unionisation bid to end use of their technology by Israel and the US military.

The UK-based Google DeepMind employees - who aim to become the first frontier artificial intelligence lab worldwide to unionise - sent a letter on Tuesday requesting management recognise the Communication Workers Union (CWU) and Unite the Union as their official representatives.

“This is a really important moment where tech workers at Google’s frontier AI lab are connecting with some of the most oppressed people in communities around the world in meaningful ways, based on foundational values of solidarity and trade unionism.”

“By exercising their rights to collectivise they are in a strong position to demand their employer stop circling the ethical drain of military-industrial contracts, echoing the sentiment of many working people in the UK and elsewhere.”

John Chadfield, CWU National Officer for Tech

Google employees have for years protested the ethics of contracts like Project Nimbus, a joint programme with Amazon making cloud computing and AI tools available to Israel, including amid the Gaza genocide. Maven, a US government project from which Google withdrew in 2019 after staff uproar, has reportedly been used in targeting in the Iran war.

The unionising DeepMind workers are seeking an end to use of Google AI by Israel and the US military. Their demands also include restoring a scrapped commitment not to make AI weapons or surveillance tools, the creation of an independent ethics oversight body, and the individual right to refuse to contribute to projects on moral grounds.

“We don’t want our AI models to assist in violations of international law, but they already are aiding Israel’s genocide of Palestinians.”

“Even if our work is only used for administrative purposes, as leadership has repeatedly told us, it is still helping make genocide cheaper, faster, and more efficient. That must end immediately, as must harm to Iranians and human lives anywhere.”

a worker at Google DeepMind

Google recently agreed to let the US Department of Defense use its AI models for classified work, a move opposed by over 600 employees. Google staff worry how the technology will be used given the Pentagon imposed restrictions on competitor Anthropic over its refusal to permit use of AI system Claude in fully autonomous deadly weapons or domestic mass surveillance.

The British unionisation push is part of a wider campaign. DeepMind employees globally are considering in-person protests and “research strikes”, where they abstain from work expected to significantly improve core products such as the Gemini AI assistant.

The unionisation bid would secure representation on pay, hours, and holidays for at least 1,000 staff tied to Google DeepMind’s London office. The employees’ letter gave management 10 working days to voluntarily recognise the CWU and Unite, or take other steps like agreeing to mediated negotiations, before the matter is escalated to a formal legal process to force recognition.

“I hope that recourse to the statutory procedure will not prove necessary.”

“We look forward to working with you in a spirit of co-operation on behalf of the workforce.”

John Chadfield, CWU National Officer for Tech

“When artificial intelligence is developed without any accountability over its direction or purpose, it is bad for society as a whole - not for just the workers involved.”

“These conscientious workers demanding greater say over what they produce with their considerable skills are acting in the tradition of all those who have fought for workers’ rights before them.”

Dave Ward, CWU General Secretary

In the press

The Guardian: Google DeepMind in talks with UK unions amid staff concern over US and Israel's AI use ↗

auth: The Guardian date: 20-May-2026 cat: Press cmpgn: Google tags: Press

Novara Media: Google AI Workers in UK Vote to Unionise Amid Deal With US Military ↗

auth: Novara Media date: 06-May-2026 cat: Press cmpgn: Google tags: Press

WIRED: Google DeepMind Workers Vote to Unionize Over Military AI Deals ↗

auth: WIRED date: 05-May-2026 cat: Press cmpgn: Google tags: Press

Show more

What is AI?

ok so verrrry broad terms about what AI is…

First, I have a short-ish version of what’s going on.

Then a longer version if you want a rough idea of how it’s actually working (explained by someone who scraped by in their undergrad degree lol)

the short version

“AI” is usually used to refer to “LLMs” - large language models. It’s a big mathematical model that takes in a large amount of text, converts it to “tokens”, and uses that to predict a result, also as text. (look there’s also images and now video butttt… it kinda does the same thing under the hood. it’s all “tokens”)

Each time an LLM is run, it works on a huge amount of calculations based on probabilities. These probabilities are worked out using training. This is a mostly automatic process where you have some input data which you know the result for, and you keep running the model over and over again and changing values until the results are close to what you expect.

Training and its energy impact

This training uses a lot of energy and data. Latest models are scouring the entire Internet, Wikipedia, textbooks, you name it, to find example data that they can use to train the models to guess the word that comes next. GPUs work quite well for this. So tech companies have been buying them all up…

Are they giving attribution or asking for permission…? Er… no. Usually not.

Running the model each time, even once the training is done, still requires loads of data, storage and energy. The environmental impact of the data usage means fossil fuel power plants are being brought back online. Tech companies are building miniature nuclear power plants just to keep up with the energy demand (small modular reactors. Data centers are stealing polluting water from surrounding land to keep their servers cool. We are going backwards on global emissions goals.

How many “r"s in “strawberry”?

Now remember I mentioned this is all based on probabilities? You betcha. This means that the output is never guaranteed. It makes mistakes, or “hallucinates”. The same input doesn’t always return the same thing. This means that LLM models can be unpredictable and work in unexpected ways. E.g., asking a model to add a list of numbers. Earlier models would convert the numbers to tokens, and because it sees something like “add these numbers” it will try to predict the output based on what it’s previously seen to come up next. This is why some models can’t even count the number of “r"s in “strawberry”.

An example. Asking GPT-4o to add up numbers.

Above: The correct answer, produced by Wolfram Alpha

Above: GPT-4o mini’s attempt.

Other models will predict that, say, there’s a list of numbers to add up, and then call upon a program to sum them before giving it back to the model to return as an answer, for instance. This is how a lot of modern LLMs now work, they are able to access a range of tools and programs to work in multiple steps to predict an answer. This is why, e.g. when you run Notion AI, you can see it print out a load of information as this process unfolds. It’s often known as “multi-stage reasoning”.

Prompts and prompt injection

A “prompt” is another word for the text input someone gives to an LLM. It can be a question, instructions and can get quite detailed.

Prompts can often be saved, used and shared as a way of encouraging an LLM to work in a certain way or provide a certain type of answer. You often hear these referred to as “skills”, “gems”, “custom GPTs”, that kind of thing. For instance, you might have one that says something like source:

Everything is fed as input into the LLM in the same way and converted into tokens the same. The chat history and the prompt used to try and control the behaviour of the model. It’s all treated the same, although by trying to separate separate bits of information like humans do in their writing (horizontal lines, headings, spaces) can try to influence these. Using strong, imperative language and even ALL CAPS can try to steer the model further into the behaviour the author wants.

Since the chat history and the original prompt are both treated the same way, does that mean the person sending messages to the customer support could use the same techniques to get the model to do what they want, even if the author didn’t want this?

…yeeeessss.

This is called prompt injection and it’s basically unfixable because of how these models operate (source). It’s been used to send malicious calendar invites to steal private information with Gemini. It’s an incredibly difficult problem!

Other bits

So there’s also the idea that because LLMs are trained on publicly available data, and more data online is being produced by LLMs, there’s this weird phenomenon where these models are training on themselves and converging towards “average” results. The Claude Opus 4.6 model has a reduced performance compared to 4.5. Generated images are increasingly becoming streaked with piss yellow because everyone in 2015 was using the X-Pro 2 filter on Instagram. “Model collapse” is something that some people believe to be an inevitability. I mean. People are sharing techniques for trying to get rid of the tint:

But also it means that when LLMs “hallucinate”, these can find their way into articles and blog posts. This finds its way into training data and now these fake statements become reinforced as “facts” according to future LLMs. It’s affecting journalism and resulted in an editor being fired. It’s alarming quite how much of the Internet is now being polluted with lies and slop.

The companies behind LLMs are morally bankrupt and are subsidised heavily by private equity. A recent Forbes article suggests Anthropic could be running up to $5,000 of computing power for a user spending $200… A lot of OpenAI’s funding is reliant on it “achieving AGI” (that’s Artificial General Intelligence, just a buzzword for a model that is smarter than a human) within a certain timeframe and it’s just not getting there. Ed Zitron has a fascinating and long article that dives into the OpenAI financials.

They are complicit in the current wars, despite the press releases. Claude has been used by the department of defence to help select targets for bombing. OpenAI seem to have no qualms about automated killbots (until they were caught). They are evil companies run by evil people.

Ok so that’s kinda about it off the top of my head. Below is a more in-depth explanation of how LLMs work if you’re interested.

(neural networks) what came first…

so like, in computing there’s a thing called a neural network. It’s essentially a program that takes some kind of inputs (like numbers!) and then does a bunch of calculations on them, combining and splitting results multiple times, to get an output (could be a number, a true/false, anything like that).

a lot of those calculations work on probabilities, so like, a certain calculation might have an 80% chance of returning a large value over a smaller one. stuff like that. these probabilities are worked out by taking a bunch of input data where we know the answer we want, and running this model over and over again, tweaking these probabilities each time to get the results closer and closer to those answers. this is what training is!

This happens on a huuuge scale. There’s many different calculations running (1000s to even millions!), a large number of inputs, and results being passed to further layers in the system. you could think of it as bit like this:

Where each of those circles is some kind of calculation (a neuron!) and each line is some kind of value being passed between it.

Even at a small scale like this, it can be really hard to unpick and work out why one set of inputs leads to an output. The idea is that it works in the same way a brain does. Lots of neurons firing and sending information around to return something that is correct, trained thousands or millions of times with learning data. (“reinforcement learning”).

You can use this for all sorts of cool things! If you imagine images as just a set of pixels with a number that represents its colour, a picture quickly just becomes a list of numbers. Standardise the data, feed it into a network, see if what comes out matches what you expect. Change the probabilities and run it again and see if it gets closer.

anndddd boom. You’ve just created something that can recognise text in images…

this powers pretty much everything in the world. tiktok algorithms, “dynamic pricing” in restaurants, shipping routes, speech to text on your phone. even on iphones, there’s a hidden model that runs on the keyboard that will grow and shrink the keys depending on what you type so you make fewer typus!

ok so now for AI

in the early 2010s, some computer scientists came up with this idea of a “transformer model”. it’s the same kind of thing except that it not only runs multiple networks at the same time, but can share results between them and look back and forwards across the data it receives. it’s now able to remember stuff. this is a “context”, which is a word that comes up so often in AI lol. but it means that one model is able to look at another model to figure out the “context” of what it’s currently looking at.

it makes training a lot easier since there’s no weird loops happening inside the model which for reasons I won’t go into, makes training faster. google dropped a paper at some point called “attention is all you need” and that was the hot shit at the time. this whole idea of being able to figure out the context of a bit of data meant that they got a lot better at being able to ‘understand’ words, language, things. it meant that a model that looks at images can figure out objects and shapes.

“AI” when its talked about at the moment usually refers to a “Large Language Model” or LLM. It’s a model operating on a huge amount of input data - human language. Transformer models are able to take a stream of text and break it down into words and phrases. Older neural networks would just treat it all as a big list of letters. This means that the model is able to simulate “reasoning” across sentences and paragraphs. The “memory” it keeps means that if a “cool cat” was mentioned at the start of the paragraph, it would be “remembered” later on when the LLM is processing a phrase about “the animal”.

This is really how a chatbot like ChatGPT or Gemini works in a basic level! When you send a message, your message and all of the messages in the chat history are turned into a big block of text and sent into an LLM. The LLM will then processes all of the text, turning it into tokens along the way, and then predicting the “most likely” answer.

Images, voice, it all works the same way. Everything is converted to tokens and run through a model and an answer is predicted.

This works okay for things like categorising data, converting speech to text, and picking out themes. It works to a slightly lesser extent for summarising large amounts of text, since its predictions are based on how similar sets of tokens (how similar pieces of writing) in its training were summarised. It can detect common themes or significant points, but can often struggle with prioritising what’s actually important. Because it’s not thinking, it’s predicting based on the data it’s been trained on.

This is why things like Gemini meeting summaries will often focus on small and insignificant things, like “the team discussed their morning commutes” over something like “the team raised concerns about the project’s ethics”. It’s all prediction, and it’s all based on probability. So there’s a chance it will do something unexpected and incorrect, and most often when you’re doing something new.

…or, how big is the carbon footprint of your AI endeavours?

The AI hype machine rumbles ever onward, engulfing the UK government and spawning a thousand experts along the way. But it’s not only good for contributing to the zombie-fication of the internet, AI is also quickly playing a starring role in climate breakdown too! There’s a growing movement flagging the significant environmental costs of developing, training, and running models, particularly the incredible resources required to maintain the data centres associated with the largest models.

The first step for most organisations or individuals concerned about the impact of their forays into AI is to understand how bad it currently is. Baselines then provide a benchmark for looking at how effective their strategies for reducing it are. With this in mind, we’ve rounded up some of the available carbon footprint calculators below, to help you take that first step.

Most calculators are based broadly speaking on the type of hardware you’re using, how long you’ll be using it for, and **where the compute power is located that you’ll be using. There are variations on this theme across all the different tools.

A quick note: the tools below all share a limitation that they are predominantly focused on the release of carbon dioxide into earths atmosphere. While this is a key greenhouse gas driving global warming, there are other emissions associated with global warming that are not included. The tools also don’t cover the water usage of data centers, another significant climate impact which will only be compounded as the climate warms further. More on the water usage of AI here

We also see a split in enterprise vs individual project level estimates: project-level estimates typically calculate the one off emissions of training a model, while enterprise tools allow ongoing monitoring of the impacts of using a model as well.

Enterprise solutions for cloud compute carbon impact

You might be surprised to know that the three major cloud service providers offer a variety of different tools to estimate the carbon emissions of customers

Microsoft Emissions Impact Dashboard for Azure

Designed not just for your machine learning models, this PowerBI app calculates the carbon footprint of your cloud-based computing in Azure.

It’s based on a Microsoft methodology validated by Standford in 2018 and includes emissions from Microsoft vendors and suppliers, as well as regional variations in fuel usage, “in line with ISOs”. Although this is Microsoft marking it’s own homework, and it only provides estimates, if you’re already using Azure, this tool should help you track and potentially reduce, the emissions associated with cloud usage.

Link: https://appsource.microsoft.com/en-us/product/power-bi/coi-sustainability.emissions_impact_dashboard

Google Cloud Carbon Footprint Dashboard

Similar to Microsoft, Google will provide you with a lovely visual dashboard for tracking how your IT cloud operations are contributing to climate breakdown. Perfect for management of multiple projects across an organisation!

The biggest surface difference between the two is that Google claims to adhere to the Greenhouse Gas Protocol for emissions reporting instead of an unknown ISO. The documentation for the methodology is also easily accessible online, vs the random PDF file from Microsoft.

Amazon Web Services Carbon Dashboard

Not to be outdone, Amazon claims to meet Greenhouse Gas Protocol AND ISO14064 with it’s carbon footprint estimates for Amazon Web Service customers. Notably however, it only includes on-site fossil fuels, and Amazon Web Service products electricity use in it’s calculations. Google, in contrast incorporates a third category (the terminology used in the documentation is “scopes”) which includes upstream emissions from data center equipment and buildings. So, while AWS might look like a lower carbon option, that’s probably not the full picture.

Essentially though, which of the above you use depends on the enterprise cloud provider your organisation has gone with. They each have their benefits and drawbacks but most of them will give you some kind of overview that can be used to make management decisions about project cloud resources and use. If they’re implmented before your company dives head first into the shallow end that is coporate AI adoption, you could also use them to demonstrate further what a bad idea that is climatically.

Individual Project Tools

Looking for tools at a different scale to the enterprise dashboards? There are a variety of different tools out there, mostly with some flavour of open license, that will integrate into your python project to provide tailored estimates

ML Co2 Impact

An open source (MIT license) project with a quick and easy website for calculating an estimate of carbon impact across cloud providers and hardwares - just select from the drop down options and hit the big red button. The data used to support all the calculations is on GitHub in case you’d like to interrogate it or add to it yourself, and the assumptions used to build the calculations are laid out clearly. There’s also a python package to install if you’d like to integrate carbon estimations in your own coding workflow (say, alongside PyTorch, for example).

Bonus points for incorporating a LateX template for the researchers looking to include calculations and providing some tips on how to reduce the carbon impact of machine learning. Downsides include difficulty in monitoring emissions over time and at a scale larger than individual projects or work.

Website: https://mlco2.github.io/impact/ GitHub: https://github.com/mlco2/impact

CarbonTracker

Another open source (MIT license) python package developed by students in Denmark, this tool will estimate the carbon emissions for training your machine learning model, and even gives you the option to end training cycles early if the carbon emission predictions are too high. It’ll automatically detect your location based on IP, then use that for part of the calculations of emissions.

CarbonTracker has been widely cited in academic papers (400+ times and counting) but as with codecarbon, it’s not clear how well it plays with enterprise perspectives and demands. This is understandable, since it was developed by PhD students, but it’s worth bearing in mind that you’ll need to do some aggregation and parsing if you want a nice dashboard for your manager to point at in a meeting. The documentation also isn’t great.

Website: https://carbontracker.info/ GitHub: https://github.com/lfwa/carbontracker

Want to get really nerdy about the tools mentioned above? “How to estimate carbon footprint when training deep learning models? A guide and review” (Bouza, Aurelie, & Lannelongue, 2023) is for you! It looks at seven different tools available in depth, so if the two mentioned above aren’t quite what you were looking for, this paper probably has one that’s a better fit. DOI: 10.1088/2515-7620/acf81b

Other options

For the non-technical folks out there, Deloitte has hopped on the AI bandwaggon with it’s very user friendly AI Carbon Footprint Calculator. Is there a rigorous methodology described? Don’t be silly, it’s a big 4 consulting firm! It uses the latest weights and metrics, they promise. What it does come with are some limited multiple choice options in an easy to read format and a rating out of 10 at the end. This writer couldn’t manage to get it give them a gold star as well, but you might have better luck. You also have the chance to enter their marketing funnel by downloading a personalised report with “recommendations”.

This is honestly, probably best for senior managers or people trying to convince senior managers that using live streamed cloud data to develop autonomous flying cars with natural language processing might be a bad idea.

So once you’ve identified the carbon impact of your project, organisation, or idea to do weather prediction with chatGPT, what to do next? We’ll cover this in a future article (Step 1: STOP PUTTING AI IN EVERYTHING), but in the meantime you can send your thoughts to contact@utaw.tech.

At UTAW, members are are organising and campaigning on issues ranging from tech complicity in genocide, employee surveillance and workplace discrimination, to climate breakdown. Join today and you too can get stuck into pushing for change across tech industry.