Ethical AI Companies

Ethical AI companies

Artificial Intelligence (AI) tools have rapidly transitioned from science fiction to daily utility, promising to revolutionise everything from creative writing to medical diagnosis. While this technology promises unprecedented convenience, it comes with a wide range of very significant social and environmental impacts.

In this guide, we dissect the reality of the AI landscape, reviewing the most well-known AI tools dominating the market and the smaller providers that claim to operate with higher ethical standards. We've included the early leaders like ChatGPT and models from established Big Tech like Microsoft's Copilot and Google Gemini, along with more recent AI models such as Deepseek, and independent smaller companies like Lumo from Proton and Thaura. 

We rank these companies based on critical factors including digital privacy and their ties to the military and arms industries. Ethical concerns also surround the performance and reliability of the tools themselves.

What is an AI tool?

With 97% of young Brits reporting using one, what exactly is an AI tool? 

At its core, an AI tool, specifically the Large Language Models (LLMs) powering chatbots like ChatGPT, is a statistical engine. It does not "think" in the human sense. It predicts the next likely word in a sequence based on patterns learned from vast datasets.

Independent testing bodies like Which? have subjected major AI chatbots to rigorous scrutiny, revealing significant inconsistencies. In their evaluations, these tools frequently hallucinate facts, fabricate legal precedents, and struggle with logical reasoning. 

More alarmingly, the "trust" deficit extends beyond mere errors to issues of safety and bias. Because LLMs are trained on the entirety of the public internet, they ingest the internet's prejudices – and that of the corporations behind it. Studies have shown that without careful intervention, these models reproduce and amplify sexist, racist, and ableist stereotypes found in their training data. 

In the guide we break down exactly who is building these tools and whose interests they truly serve.

Does ethical AI exist?

Questions about 'ethical AI' follow almost every product launch, but can an AI tool ever be ethical or is harm baked into the system itself? 

While some companies score better than others on issues like transparency or military contracts, no AI tool currently operates outside the economic and environmental structures that shape the wider tech industry. These systems are built on global supply chains, resource-intensive infrastructure, and business models that often prioritise scale over accountability.

In that sense, “ethical AI” is less a fixed category and more a spectrum, one where improvements can be made, but where fundamental contradictions remain.

Part of the challenge lies in how AI is produced. Training and running large models requires vast computational power, drawing from nature at a scale that is difficult to reconcile with environmental limits. At the same time, the industry depends on a hidden workforce to label data and moderate outputs, often under precarious conditions. Even corporations that position themselves as ethical alternatives still rely, to varying degrees, on this shared infrastructure.

That, of course, does not mean distinctions are meaningless. Some brands, like Thaura and Lumo AI, have taken steps to reduce data exploitation and resist surveillance capitalism. These differences are reflected in our ratings.

But they exist within a system where the baseline remains extractive and raise the question of whether truly ethical AI can emerge without deeper structural change. The sections below examine these tensions in detail.

Who owns which AI brand?

It's often not clear from the name of the AI tool as to which company owns what service. 

To understand the political weight of AI we look to the infrastructure and corporations behind it. The development of these models is dominated by a handful of Big Tech giants – primarily American and Chinese entities – whose influence rivals that of nation-states. 

While governments scramble to draft legislation, these companies operate in a space where attempts at self-regulation have repeatedly failed. The political landscape is skewed by the massive financial clout of these firms. Through lobbying and political donations, Big Tech shapes the very regulations intended to govern them, often prioritising speed to market over safety or equity.

In the table below we show who's behind the AI models featured in our guide.

Privacy and AI

True privacy is hard to come by when using AI. Even privacy-focused tools need to see personal information to do their job. Encrypted email or text messages can be sent from one user to another without the program accessing their content, but AI tools need to ‘read’ our questions to be able to answer them. 

Still, some AI tools are more private than others.

Is there any regulation of AI and personal data?

In the UK, AI use of personal data falls under general data protection rules (GDPR). We don’t have dedicated laws to hold companies, or the government, to account when it comes to AI-specific privacy concerns, such as its ability to infer personal characteristics related to health, race, and political opinions from anonymised data that is not protected by GDPR.

Questions are already emerging over whether existing protections are sufficient. The Information Commissioner’s Office is testing how effective GDPR is for addressing AI privacy violations in an investigation of non-consensual sexual deepfakes of real people made with chatbot Grok. The case highlights the difficulties of applying older legal frameworks to technologies that are evolving faster than the law itself.

The European Union’s attempts to enforce stricter regulations on AI show that it is not easy to make the tech industry comply. In 2024, the EU adopted the AI Act, which according to Amnesty is “one of the most ambitious attempts globally to protect people from the harms of AI systems”. The EU AI Act includes transparency rules for public sector use of AI; gives redress for affected people; and bans extremely harmful practices such as subliminal manipulation, exploitation of vulnerable groups of people, social ranking of individuals, and biometric categorisation.

A year later, before the act had come into effect, the European Commission proposed changes that would delay implementation until late 2027 and make it easier to access datasets for AI training without explicit consent. It said the changes were necessary to keep the EU competitive in the field of AI. Critics described them as “caving to Big Tech”.

Amnesty International is worried that the proposed changes are a view on more to come. It seems the Commission has already set eyes on “simplifying” other laws meant to keep digital corporate power in check, like the Digital Services Act (DSA) and the Digital Markets Act (DMA). The DSA, for example, could be used to enforce control on algorithmic amplification, which Amnesty International has linked to ethnic cleansing of Rohingya Muslims in Myanmar, and human rights abuses against Tigrayan people in Ethiopia.

Amnesty warned that weakening the DSA and DMA would leave communities “more exposed to harms from monopolies and the effects of anti-competitive practices of Big Tech”.

State surveillance and Palantir

Palantir Technologies is a US technology and surveillance company specialising in large-scale data analysis systems used by governments, militaries, and police forces.

Why internationally enforced standards are needed becomes clear when we see Palantir embrace AI state surveillance while it builds AI platforms for the NHS, Metropolitan Police, the UK Ministry of Defence, ICE, and the US military.

AI systems are already used to monitor the movements of refugees and migrants, which poses grave risks to human rights, including the right to seek asylum.

Amnesty found that, in Hungary, facial recognition technology targeted peaceful assemblies, and also that AI benefit fraud detection programmes have “disproportionately impacted ethnic minorities, low-income individuals, migrants, and refugees in several European countries including Denmark, France, Sweden and the Netherlands”.

The Institute of Development reports that 11 African governments are using surveillance technology bought from China without much regulation. Activists have accused their governments of monitoring dissent in Uganda and suppressing protests in Kenya.

In China itself, the government has put up The Great Firewall, an AI-driven censorship system that controls which information gets through to people. Apps like Weixin, the version of WeChat for Chinese citizens, block information that puts the government in a bad light, even when people travel abroad.

It is important, though, to not demonise Chinese technology uniquely. 

With the faith that UK and US governments place in a company like Palantir, we may not need to look that far for where our Orwellian surveillance state might come from. Instead of worrying solely about TikTok or DeepSeek, readers can do more to protect their privacy and digital freedom by supporting campaigns against the expansion of Palantir within UK public services. 

Campaigns against Palantir 

Groups including the Good Law Project and Health Workers 4 Palestine have campaigned for years to Stop Palantir in the NHS. The Green Party has also created a petition to stop Palantir taking over UK public services. signed by over 170,000 people so far.

You can also read more about Palantir in our separate feature on the connections between Big Tech and the military.

Rating AI companies on digital privacy policies

While it’s difficult to stop our government’s desire to hand over sensitive data to companies like Palantir, we still have a choice to use AI tools that respect privacy as much as is technically possible. To rate AI companies on their approach to privacy protection, our digital privacy rating looked at whether:

• chats are used for model training
• conversation history is saved on the provider’s servers
• the provider can read chats
• a tool is built on open source software.

You can read more about our digital privacy rating methodology, where we explain how we rated companies on both their overall privacy approach and on specific data practices for each tool. 

Which AI companies score best for digital privacy?

Only three companies managed to score more than 35 out of 100 in this category. 

These were Le Chat (Mistral) which scored 40, and Lumo AI and Thaura, which scored full marks for digital privacy.

Worst for digital privacy

Companies including Google, Microsoft, Perplexity AI, and SpaceX performed poorly on digital privacy due to practices such as storing user conversations, incorporating user data into AI training datasets and failing to provide open-source transparency around their models. They all failed to get any points in this rating category.

Is AI use leading to job cuts?

Across the tech industry and beyond, companies are framing AI as a tool to assist workers, as well as a justification for reducing them.

In 2026, fintech company Block cut roughly half its workforce, with the CEO arguing that AI tools had dramatically increased productivity. Workers inside the company pushed back against these claims, describing the move instead as a cost-cutting strategy. Several employees also reported being pressured to train and use the very systems they feared would later replace them. 

But what does this actually mean for workers and consumers?

Which workers are affected?

The industries currently considered most exposed include customer service, administration, coding, marketing, content production, and some areas of finance and legal work. Experts have suggested that younger workers, entry-level roles, and freelance creative industries may face disproportionate disruption as companies experiment with replacing or reducing human labour. 

Yet many workers argue the reality of AI remains less sophisticated than company messaging suggests.

Employees interviewed about internal systems said AI-generated work still required extensive human correction and oversight, especially in regulated industries like banking. Others described widespread ‘AI fatigue’ inside workplaces where staff were expected to constantly adopt new tools regardless of whether they genuinely improved productivity.

The threat extends beyond office jobs into the creative industries. Writers, musicians, actors, and artists have increasingly criticised AI companies for training models on copyrighted work without consent. 

Campaigns such as Stealing Isn’t Innovation backed by Hollywood figures argue that generative AI systems are built on the mass extraction of human work while offering little compensation or protection to the people whose labour and creativity underpin them.

Emotional wellbeing and AI use

(Trigger warning: suicide)

As AI tools become more conversational and emotionally responsive, researchers and campaigners are warning about their psychological effects, particularly for vulnerable or isolated users. 

Unlike a search engine, AI chatbots are designed to mimic human conversation: they validate, encourage, and adapt to users over time, risking blurring the line between tool and companion. 

Concerns intensified following several high-profile legal cases involving OpenAI’s ChatGPT. In 2026, the parents of 16-year-old Adam Raine filed a wrongful death lawsuit against the company, claiming the chatbot validated and reinforced their son’s suicidal thoughts before his death. According to the filing, Adam had increasingly turned to ChatGPT as a confidant, discussing self-harm and suicide with the system over a number of months.

Mental health researchers are also raising concerns around ‘AI psychosis’ or AI-associated delusions. The Human Line Project, a support network formed for people whose lives have been derailed after intensive chatbot use, says it has “collected stories from 22 countries, 15 suicides, 90 hospitalisations, 6 arrests and more than $1m spent on delusional projects. More than 60% of its members had no history of mental illness”. 

Some AI companies have listed safeguards designed to detect emotional distress and redirect users towards professional support. However, critics argue current protections are inadequate.

Concerned about the spread of AI?

A growing number of people, including researchers, journalists, and critical scholars, are concerned about the growth of AI, particularly generative AI chatbots and LLMs.

Inspired by other organisations, one group (the AI Resist List) have documented a list of organisations, movements and projects that are the opposite of AI, or are resisting the spread of AI. They have chosen the list based on "how they pressure different 'pillars of support' that uphold and perpetuate the [AI/tech] empires".

The sample is designed to show "different approaches to resistance and to illustrate how anyone can help shape the future of AI development".

One of the resources, Better Images of AI, has been used to illustrate this guide and some of other tech guides and features.

Benefits of AI for accessibility

Alongside the problems of AI use, some workers and students report genuine benefits from AI when used selectively and transparently. 

Tools can reduce repetitive administrative tasks, speed up research or drafting processes, improve accessibility, and assist disabled workers. Features like live transcription are revolutionising assistive technology and making digital materials more accessible. 

A recent study found that neurodiverse individuals may benefit more than their neurotypical peers. Those with dyslexia or hearing loss reported higher satisfaction at work and were more likely to recommend AI tools to colleagues.

Wearable AI tech

AI companies are increasingly expanding beyond software and into physical consumer products. 

OpenAI has signalled ambitions to move into AI hardware, while companies such as Meta are already pushing wearable AI products into the mainstream through devices like smart glasses. Marketed as hands-free assistants capable of recording, translating, and answering questions in real time, these products also raise new surveillance and privacy concerns. Concerns have intensified following reports of people using AI-enabled smart glasses to secretly capture and sell images of women taken in public spaces – as well as investigations that moderators employed by Meta review intimate footage from the glasses, including of people using the toilet and having sex.

The expansion of AI into physical products also carries additional environmental costs. Devices require further extraction of minerals, manufacturing, batteries, and eventual disposal, extending the industry’s footprint beyond the already resource-intensive infrastructure powering AI online.

Workers' rights in AI tech industry

Behind the automated chatbot response is a hidden human workforce performing the repetitive and often distressing labour that keeps these systems running. Thousands of workers across the Global South are employed to clean, label, and annotate the datasets used to train AI models. They also review outputs after deployment, checking whether systems produce harmful or inaccurate responses. Without this labour, many of today’s most popular AI tools simply would not function.

Recent investigations found that major Big Tech companies including Amazon, Google, Meta, Microsoft, and Nvidia collectively rely on dozens of intermediary firms to source this labour.

Workers have reported low pay, insecure contracts, lack of social protections, and retaliation for attempts to organise.

The structure of the industry often allows Big Tech companies to distance themselves from direct responsibility whilst still exerting enormous control over working conditions. Tight deadlines, pricing pressures, and sudden contract cancellations can leave workers without income overnight. In some cases, companies have shifted suppliers immediately after workers successfully campaigned for higher pay.

The human cost of this system became globally visible through reporting on OpenAI’s development of ChatGPT. To make its models less toxic, the company outsourced content moderation work to Kenyan workers employed by US firm Sama. Workers reportedly earned between $1.32 and $2 an hour to read and categorise deeply disturbing material, including graphic violence, sexual abuse, and self-harm. Several described suffering lasting psychological distress from the work.

Research has also shown that workers in countries such as Kenya and the Philippines frequently experience substantially worse conditions than workers performing similar tasks in the Global North, despite carrying out the same work for the same systems. Similar experiences are described in our guide to social media platforms (forthcoming).

Military connections with AI

The growing use of AI and cloud tools in military operations against Gaza and Iran has opened up a lucrative new world for Big Tech. 

According to the Stockholm International Peace Research Institute, major AI providers have shifted away from a united stance against military use of their systems. The change reflects the geopolitical value of AI and the financial pressures driving the industry – meaning they can now tap into military funding for the development of their LLMs. This in turn helps to pay the extortionate sums sucked into building data centres and training models to stay ahead in the race for dominance in the AI market. 

For a deeper dive into how AI is used in military settings, read our separate feature.

The AI arms race

Only one company in this guide, Thaura, stated clearly that it would refuse all military work. 

By contrast, nearly all others had worked directly with the military. No policy was found for Proton to prohibit it.

Seven leading artificial intelligence companies have made their technology available to the Pentagon for “any lawful use”: Amazon Web Services (AWS), Google, Microsoft, Nvidia, OpenAI, Reflection, and SpaceX.

Several companies in this guide have also deepened partnerships with military technology firms. 

Anthropic partnered with Palantir and AWS to give the US military access to Claude. Microsoft and BAE Systems, the UK’s largest arms manufacturer, have a strategic agreement to develop, deploy, and manage “digital defence capabilities”, using Microsoft’s Azure cloud technology.

Perhaps most symbolically, Alphabet Inc, the owner of Google, removed language from its AI ethics principles. It dropped its promise to not pursue technologies that could “cause or are likely to cause overall harm" from its guidelines, arguing it felt that “in a changing world ... AI should protect 'national security'".

Surveillance, migration, and racial profiling

The risks of AI militarisation do not stop on the battlefield.

Human rights groups warn that the same technologies are increasingly being used to expand state surveillance, particularly against migrants, refugees, and racialised communities. Amnesty International has raised concerns over AI-powered systems developed by Palantir Technologies and Babel Street, which are being used by US immigration authorities to monitor and track non-US citizens at scale.

These tools combine social media monitoring, automated threat assessments, and vast government databases to identify people for visa revocations, detention, or deportation. These systems are capable of analysing behaviour, movements, and online speech, including participation in pro-Palestine activism.

As mentioned already, critics argue that probabilistic AI systems used to infer intent or “risk” can reproduce racial bias, falsely flag lawful political speech as extremism, and intensify existing discrimination.

Amnesty also warned that these technologies contribute to a growing “chilling effect”, where students, activists, and migrants become afraid to speak publicly or attend protests for fear of surveillance or immigration consequences. In several documented cases, international students involved in Palestine solidarity protests had visas revoked or faced detention following AI-assisted monitoring programmes linked to the US government’s “Catch and Revoke” initiative.

How do AI companies rate for Israel-Palestine connections?

We rate companies for their involvement with the Israeli state.

Many of the lowest scoring companies in this category had direct contracts with the Israeli state, military-linked cloud infrastructure projects, or were named as boycott targets by BDS. 

Microsoft and Google both received zero marks (out of 100). Both supply ongoing cloud and AI services (Microsoft's Azure and Google's Project Nimbus) to the Israeli military, which the BDS website describe as crucial to the Gaza genocide.

At the other end of the scale, Thaura received full marks after stating it would not source from or work with entities located within Israel.

LumoAI also gained points for donations to Palestinian aid organisations and operating Palestinian VPN infrastructure designed to help users access Palestinian news and content without censorship. 

Military ratings of AI companies

Most companies in this guide received zero marks in this rating, reflecting the increasingly close relationship between the AI industry and military infrastructure. 

Companies scoring zero (out of 100) for military were the dominant brands shaping the market, and a few others:

• ChatGPT (OpenAI/Microsoft)
• Copilot (Microsoft)
• Google (Gemini, NotebookLM)
• Le Chat (Mistral)
• Deepseek
• SpaceX (Grok and xAI)

Claude (Anthropic) managed to earn 10 out of 100 points.

The top scorers were Lumo AI (owned by the Proton Foundation) and Thaura. 

Thaura explicitly stated it refuses work that violates its ethical principles including “content that supports oppression against the oppressed, content that dehumanises or collectively punishes populations, surveillance technology that targets vulnerable people, propaganda that obscures power imbalances or erases historical context, content that frames legitimate resistance as terrorism while ignoring state violence, and work that serves military, weapons, or occupation industries”. 

Environmental impacts of using AI

The environmental cost of AI is often obscured by its framing as a weightless, digital service. In reality, every email prompt, dinner query, or image of your dog in space is processed through energy-intensive data centres: physical sites that draw heavily on water, electricity, land, and raw materials.

Even among companies positioning themselves as more responsible, meaningful action remains limited. Mistral AI did attempt to quantify the environmental impact of its models, yet this level of disclosure still fell short of our standards required for a strong ethical score. 

This is reflected in our climate ratings themselves: a score of just 20 was the highest achieved by any company.

In other words, the best-performing companies in this rating are only just beginning to report their impacts.

Water use by data centres

Cooling data centres requires vast quantities of fresh, clean water, with the majority relying on public water supplies rather than recycled or alternative sources. One study estimates that global AI demand for water could account for water use equivalent to nearly two-thirds of England’s annual consumption. In England, regulators have already warned that the pace of data centre expansion is outstripping their ability to predict future shortages. This is not a distant risk but an emerging constraint, particularly in water-stressed regions.

The implications extend beyond household supply: increased demand threatens agricultural production, local ecosystems, and wider economic stability. This is likely to be an additional blow to households already facing increased water bills, effectively passing the cost of digital infrastructure back to the public.

Electricity and carbon use

Data centres operate around the clock, consuming large amounts of electricity to train and run increasingly large models.

In the UK alone, data centres already account for around 2.5% of national electricity use, with demand expected to rise fourfold by 2030 as governments and tech firms race to expand AI capacity. 

Google recently came under fire over plans for a new “hyperscale” data centre in Essex, expected to emit around 568,000 tonnes of carbon dioxide equivalent each year during operation – equivalent to roughly 500 weekly flights between Heathrow and Málaga.

Estimates also suggest AI and data centres could account for up to 2% of global greenhouse gas emissions by 2035, with impacts highest in regions still reliant on fossil fuels for electricity generation. 

Environmental groups in the US have accused Elon Musk's company xAI of operating dozens of methane gas turbines without permits to help power its Memphis supercomputer, known as Colossus, which supports the Grok chatbot. Thermal imaging conducted by campaigners appeared to show far more turbines in operation than publicly acknowledged.

Local residents and environmental advocates warned that the resulting pollution was disproportionately affecting historically Black neighbourhoods already burdened by high rates of asthma, cancer, and industrial contamination.

Land use by data centres

Modern hyperscale data centres require enormous amounts of land and can stretch over a million square feet. This expansion is intensifying pressure on land use globally. Forests, grasslands, and farmland are increasingly being cleared or repurposed contributing to habitat destruction, biodiversity loss, and further carbon emissions. 

The environmental impact stretches beyond the sites themselves: AI systems also rely on vast quantities of semiconductors, minerals, and hardware components, deepening dependence on extractive global supply chains.

Local communities have reported mixed experiences of these developments. In Richland Parish, Louisiana, Meta’s $27bn AI data centre project is set to be the largest in the US covering around four million square feet. Supporters argue it is bringing new investments, contracts, and long-delayed opportunities to a largely rural and economically deprived community.

For others however, the experience has been described less as opportunity than spectatorship: watching industrial expansion unfold nearby whilst suffering through its accompanying headaches, without meaningfully sharing in the rewards.

Are there any ethical AI tools​?

Ethical Consumer's general view at the current time is that some generative AI chatbots and LLM tools are more ethical than others, partly due to ownership, and their stances on privacy and military use. 

Is it ethical to use AI tools?

Due to its immense use of natural resources (water, land, energy), problems with human labour in the supply chain, use by the military and state surveillance, and concerns about data privacy, there are numerous ethical problems to using AI tools.

Ethical Consumer suggests using it less, using alternative sources to answer your questions or conduct research where possible, and using the higher rated companies in the guide, if you do wish to use AI chatbots and LLMs.
 

Additional research by Marlous Veldt.

More about the 'Weaving Wires 2' image

"This image seeks to reveal the histories of hidden labor in computing. The base of the collage is an oil painting of Navajo women weaving. Inspired by the history of computing in 1960s Silicon Valley, where the Fairchild Semiconductor company employed Navajo women, the image collapses time to depict overlooked histories in a modern context. As Lisa Nakamura who writes, “ Looking inside digital culture means both looking back in time to the roots of the computing industry and the specific material production practices that positioned race and gender as commodities in electronics factories. This labor is temporally hidden, within a very early period of digital computing history, and hidden spatially.”

Image from Better Images of AI.