Why this matters

Over $250 billion was invested globally in AI development in 2024. Less than 0.1% of that went to safety research. The most powerful AI systems are being built behind closed doors, with minimal public oversight or democratic input - and public voice on the direction of AI remains thin, despite strong majority support for regulation and oversight.

What we found

A growing but scattered ecosystem of civil-society groups is beginning to challenge the direction of AI development through public voice, protest, and civic participation. The report maps four strategic clusters: protest and public mobilisation (PauseAI, Stop AI, Control AI); narrative-shaping and watchdogs (Algorithmic Justice League, DAIR, Fight for the Future); public literacy and democratic engagement (We and AI, Connected by Data, CivAI); and field-building and political lobbying (Ada Lovelace Institute, Stop Killer Robots). Within the movement, there are sharp disagreements between those focused on AI harms affecting people today – from algorithmic bias to job displacement and surveillance – and those focused on longer-term existential risks from superintelligent systems. Most activity is concentrated in North America and Europe, most groups are volunteer-run or single-issue focused, and coordination across the ecosystem remains weak.

What it means for funders and campaigners

The civic side of AI safety is small, underfunded, and loosely connected – but could be significantly amplified with targeted support. A fund committed to public-interest AI, modelled on what the Climate Emergency Fund does for climate, could support informal and innovative groups alongside established actors. Key priorities include strengthening movement infrastructure (coordination, legal support, cross-issue coalitions), challenging the dominant narrative that AI development is inevitable, and building public-facing AI safety capacity outside North America and Europe. The AI safety movement can also learn from other social movements about how to create more than the sum of its atomised parts.

Read the full report

The findings above are a summary. The full report, including methodology and supporting evidence, is available on socialchangelab.org.

Read the report
AI safety