Figure 2: Total AI funding and AI safety funding in dollars, in 2024
 

This asymmetry constitutes a huge risk. As AI companies perceive themselves locked into a winner-takes-all AI arms race, ethics and safety are subordinate to speed. Without civic mobilisation, the default is private firms shaping a general-purpose technology with world-altering consequences, unchecked by the societies that will have to live —or die—with it. Let’s not forget that less than two years have passed since the CEOs of the major AI labs signed a statement saying,“Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war”. It is time they were held to account on this. As the Ada Lovelace Institute argue, public involvement is essential “so that technology can develop in alignment, rather than to an agenda driven by economically pressured policymakers and commercially driven technology companies”.

The Scope of this Work

This report is an early-stage mapping exercise. We are not attempting a comprehensive account of all AI harms or a complete directory of every relevant group. Rather, we aim to give an overview of:

• The broad categories of risk posed by advanced AI systems—risks to whom, over what timescale, and with what likely magnitude.

• The existing ecosystem of actors working to mitigate those risks—focusing primarily on grassroots movements and campaign groups.

• The gaps that exist in terms of risk targets, strategies, geographies, and communities.

There are many different ways to categorise AI risks. Some frameworks focus on causes: malice, misalignment, misuse. Others focus on domains: discrimination, misinformation, loss of control, economic instability. Still others distinguish between near-term harms and long-term existential risks. In this report, we use a hybrid framework to clarify what’s at stake and who is working to address the threat.

Mapping the Threat Landscape

When people hear about AI risks, they often imagine something distant and abstract, often based on science fiction narratives—sentient machines (as in I, Robot), runaway algorithms (like The Matrix), or apocalypse (as in Terminator 2). But AI-related harms are already here, embedded in hiring software, recommendation engines, predictive policing, and viral disinformation. To shape an effective civic response, we need clarity on what these threats are, who they affect, and where they might lead.
 

This section offers a working map of key risk areas, structured by scale and time horizon. 
 

A working taxonomy of AI risks

We group AI risks into three broad types, adapted from the UK’s 2025 International AI Safety Report and other public-facing taxonomies:

1. Malicious use

The use of AI to cause deliberate harm, including:

• Deepfakes and disinformation

• Cyberattacks, malware, phishing

• Manipulation of elections or public discourse

• Autonomous weapons, bioweapons, other weapons to cause mass harm
   

2. Malfunctions and misalignment

The risk of AI systems behaving in unintended or harmful ways, even without malicious intent:

• Bias and discrimination in decision-making

• Errors in life-critical domains (e.g. medicine, criminal justice)

• Lack of transparency or explainability

• Misaligned goals in powerful systems
   

3. Systemic risks

Large-scale social, economic, or environmental impacts:

• Mass job displacement and labour market instability

• Erosion of democratic norms and civic trust

• Data theft, copyright infringement, threat to human creativity

• Surveillance and loss of privacy

• Resource extraction and environmental costs

• Long-term existential risks from super-intelligent system

​

These categories are porous—one threat can lead to another. For instance, mass misinformation (malicious use) can undermine trust in elections and democracy (systemic risk). A highly capable AI system that fails to act as intended (misalignment) could have catastrophic consequences at scale (systemic risk). 

​​

Some examples of how risks are already showing themselves

1. Bias and discrimination
 

One of the most well-documented real-world harms from AI is algorithmic bias. For example, in 2020, Robert Williams, a Black man in Detroit, was wrongfully arrested—in front of his children—based on a faulty facial recognition match. He spent 30 hours in jail. He’d never been to the store where the alleged crime occurred. His story is not unique; the Algorithmic Justice League has documented multiple cases where facial recognition misidentified people of colour, leading to arrests, job loss, or reputational harm.
 

These are not bugs. They are systemic failures embedded in training data which disproportionately harm marginalised groups.
 

2. Misinformation and deepfakes
 

ChatGPT recently falsely claimed that a real Norwegian man, Arve Hjalmar Holmen, who has never committed a crime, had murdered two of his children. He has filed a defamation claim. Also in 2025, AI-generated images falsely depicting riots in downtown LA circulated on social media when President Trump sent in troops to respond to anti-ICE protests. The images were shared thousands of times, fuelling conspiracy theories and outrage, before being debunked. The previous year, deepfake videos began circulating of Taylor Swift fans endorsing Trump, part of a coordinated campaign to drive voter turnout under false pretences. 
 

Of course, propaganda has long existed but AI changes the game, making it cheap, fast, convincing, and indefinitely scalable. As generative tools become easier to use, the challenge could move from detecting individual falsehoods to maintaining any shared sense of social reality.
 

3. Job displacement
 

While AI’s impact on jobs is still unfolding, early signals are clear. In 2024, a survey by the Trades Union Congress (TUC) found that 22% of UK workers had seen tasks automated by AI in the past year, often without consultation or protections. Many tech workers are already being replaced by AI and some companies like Duolingo have boasted about becoming ‘AI first’. Tech journalist Brian Merchant has documented countless personal stories of how people lost their jobs to AI. In June, research by Adzuna, the job search site found that the number of entry level jobs in the UK was down by a third since November 2022 (when ChatGPT was publicly released). Anthropic boss Dario Amodei recently said that AI could wipe out 50% of white collar jobs within five years.
 

The threat is economic, psychological and sociopolitical. As AI displaces work, it risks not only cutting off employment income, but also hiking unemployment rates and eroding people’s sense of purpose.
 

4. Loss of privacy and surveillance creep
 

AI-enabled surveillance is increasingly embedded in public life. Systems that track your face, voice, emotions, or typing patterns are now used in schools, workplaces, and public spaces, often without consent. In authoritarian regimes, such tools are already used to suppress dissent. Some Chinese citizens feel unsafe even outside of China, as they are surveilled online and at protests. According to a Wired investigation, the US Immigration and Customs Enforcement (ICE) has recently introduced facial recognition software into the phones of its officers. Even in democratic countries, surveillance capabilities are outpacing legal and institutional safeguards. Groups like Fight for the Future have campaigned—and had successes—against this creep, calling for bans on facial recognition in policing and public control over data.
 

5. Risks to humanity
 

At the far end of the risk spectrum lie scenarios that motivated much of the original AI safety field: the emergence of systems that are vastly more intelligent than humans and not aligned with our values. Although this sounds like science fiction, autonomous weapons systems (‘killer robots’) can already identify and kill individuals on the battlefield with very limited human oversight. Some argue that building systems with unprecedented intelligence and undefined applications is simply not possible to do safely. 
 

The risks of superintelligent AI are currently harder to illustrate with individual stories but potentially of much greater magnitude. The UN’s 2024 resolution on AI governance explicitly cited the need to prevent “loss of human control” over powerful systems. Groups like the Center for AI Safety, Future of Life Institute, and Control AI argue that just as with pandemics or nuclear weapons, we must act before catastrophe; leaving it too late might, at its worst, mean the end of the human race. As Control AI argues, “Humanity’s default response to risks of this magnitude should be caution. Today, the position is to allow private companies to keep racing until there is a problem, which is as untenable as allowing private companies to build nuclear reactors until one melts down.” 
 

While the probability of these outcomes is debated, their impact could be irreversible. And unlike other technologies, advanced AI could redesign itself, making course correction increasingly difficult with each new generation.

So the risks are clear, and they are legion. In the next section, we turn to the civic response. We look at the campaigners, educators, coalition builders, and movement mobilisers who are starting to take on the threats by putting the voice of the public at their heart.
 

How is Civil Society Currently Responding? Mapping the Movement

Although polling shows strong public majorities in favour of regulation, oversight, and pause measures, this has not yet translated into concerted, mass public action. Infrastructure for AI safety groups remains limited, and very few organisations are currently explicitly focused on helping people understand, track, and challenge AI deployment in their own lives and beyond. Most advocacy has been insider-focused: expert-driven, slow-moving, and disconnected from broader public energy. Now, a growing but scattered ecosystem of groups are beginning to challenge the direction of AI development—not from within labs or policy corridors, but through public voice, protest, and participation. This section maps some of the actors driving the beginnings of that shift.
 

We focus here on groups that place public engagement at their core: those informing, mobilising, representing, or giving voice to the people who are often left out of elite AI governance conversations. These are the early-stage architects of a people-powered AI safety movement. As with the risks, there are many ways to categorise these groups; academics at Bristol university group anti-AI responses as ‘Tame, Resist, Dismantle, Smash, Escape’. WeandAI suggests ‘Resist, Refuse, Reclaim, Reimagine’. There are many other taxonomies. Here, we group these organisations by their dominant strategy and how they seek to involve the public.
 

1. Protest, disruption and public mobilisation

These are groups applying civil resistance tactics to AI development—calling for pause, accountability, or bans on specific technologies.

• PauseAI
  A decentralised, international movement urging a halt to the development of frontier AI models until safety standards are in place. Known for their public demonstrations and clear, accessible communication on extinction risk.

• Stop AI
  A civil resistance group demanding a ban on artificial general intelligence. Their direct action approach reflects growing urgency and frustration at the laggardly approach of governments to AI risks.

• People vs big tech
  Innovative largely youth movement, campaigning against disinformation, social media hate, surveillance and demanding a better distributed tech ecosystem.

• Control AI
  UK based organisation ‘fighting to keep humanity in control’. 45,000+ people have signed their open statement on AI safety frontier models.
   

2. Narrative-shaping and watchdogs

These organisations expose current AI harms, challenge dominant narratives, and advocate for justice-focused AI.

• Algorithmic Justice League (AJL)
  Founded by Dr Joy Buolamwini, AJL blends research and storytelling to expose bias in AI, especially facial recognition. Their focus is on highlighting and mitigating biases in AI systems.

• Distributed AI Research Institute (DAIR)
  Led by Dr Timnit Gebru, DAIR conducts interdisciplinary AI research rooted in ethics, focusing on the societal impact of AI. They challenge the industry through rigorous, community-rooted research.

• Fight for the Future
  A veteran digital rights group using online organising, viral campaigns, and policy pressure to resist surveillance and defend civil liberties. Their AI-specific work is fairly new, but their infrastructure and reach could make them a key actor.

• AINow Institute
  A research institute confronting the power of big tech power. Their goal is to ‘reclaim public agency over the future of AI’.
   

3. Public literacy and democratic engagement

These organisations focus on informing, equipping, and involving the public in decisions about AI, to help address AI’s ‘democracy problem’.

• We and AI
  A UK volunteer-led organisation focused on increasing public understanding and engagement with AI. They offer workshops, resources, and coalition-building rooted in values of equity and inclusion.

• Connected by Data
  A UK group pushing for public involvement in AI and data governance. Their work reframes data and AI as democratic issues, not just technical ones.

• CivAI
  A California-based nonprofit that uses interactive demos—build your own deepfake, write a phishing email—to give people visceral insight into AI’s risks.

• Citizens’ Assemblies on AI / Democracy Next
  Facilitators of deliberative processes where members of the public discuss and shape AI policy recommendations. Their model offers an alternative to technocratic, big corporation decision-making.

• Global Citizens Assemblies
  Of course AI is a global rather than a national problem and some organisations argue that we need the involvement of people at the same (global) level.
   

4. Infrastructure, field-building and political lobbying

Some organisations support civic engagement, helping build connective tissue for the movement and bring the public’s voice into legislation.

• Ada Lovelace Institute
  An independent research institute with a mission to ensure “data and AI work for people and society”. Play an important convening role within the AI community and do research to influence good AI policy.

• AI Incident Database
  A searchable living archive of real-world AI harms. Like a “black box” recorder for AI failures, from deep fakes to election manipulation to surveillance—which helps activists, journalists, and researchers understand patterns of risk.

• Data & Society / Future of Life Institute / Humane Intelligence
  While not grassroots groups, these organisations contribute critical research, convening, and communication that informs public-facing efforts to keep humans at the forefront of AI developments.

• Stop Killer Robots
  A global coalition of over 250 NGOs campaigning for international law to prohibit autonomous weapons systems. 
   

Observations and Propositions from the Map

​

• The civic side is small, underfunded and loosely connected but could be amplified. Few groups operate at scale. Many are volunteer-run or single-issue focused. The ecosystem could benefit from regular events to bring all the constituent groups together to find their common overarching purpose. Funders supporting novel and innovative ways to engage and activate the public could have outsized impact.

• Geographic skew. Most activity is concentrated in North America and Europe. Efforts to build public-facing AI safety infrastructure outside these regions remain scarce and under-supported.

• There are gaps in movement infrastructure. Coordination, shared platforms, legal support, and cross-issue coalitions are all underdeveloped.

• There is dissent. For example, those who are focused on harms already being experienced by people today and those whose focus is existential risk to humanity are often not well aligned. Note that this is not uncommon in social movements. In the climate movement, for example, tech solutionist and climate justice approaches are sometimes at odds. In a movement, groups needn’t all collaborate or cooperate, but it is usually helpful to understand where other groups are coming from and for actors to understand their role in the broader movement ecology.

• Narrative power is a key battleground. The idea that “AI development is inevitable” is dominant—not surprising given the power of the tech giants. Several groups aim to challenge that, though it is currently not clear what messaging is most effective. Studies investigating which strategies are most effective would be helpful in this context.

• The AI movement can learn from other movements. Creating a movement that is more than the sum of its parts, out of several atomised organisations requires a lot of work on the interstices. Organisations like NEON who help with communications, operations and movement building, and the AYNI institute who provide training for activists and organisers are a key part of these ecosystems for other movements.

• A fund committed to ethical AI could make a real difference. The fund could support groups working outside of traditional structures (corporations, unions), by being set up to support informal and un-constituted groups with innovative approaches as well as established groups (like the Climate Emergency Fund).
   

Conclusion

There are many, highly varied risks emerging from current AI developments. Without a strong civic response, these risks present serious societal challenges. Given the nature of AI developments (fast, unpredictable, wide-ranging), social movements may be particularly well-suited to respond. They could be critical for putting the necessary pressure on tech companies and policy makers to proceed with the caution the public wants; movements could demand serious and multiple existing and emerging concerns to be properly addressed. 
 

The AI movement is aware of current threats, and is starting to mobilise against risks likely to materialise soon and in the longer term (including extinction-level threats). We hope that this report encourages more efforts in this space, giving advocates an overview of current projects and initiatives, pointing other researchers to knowledge gaps where findings could help give actionable insights to movement actors, and encouraging funders to urgently support promising approaches.