Who is building this technology, who is watching it, and who is trying to govern it.
Independent research institute ensuring data and AI work for people and society.
A London institute funded by a charitable foundation that scrutinises how AI is used in public services and how the UK regulates it. It produces evidence rather than campaigns.
Policy research on the concentration of power in the AI industry and its social consequences.
AI Now is a New York think tank that frames AI as a question of corporate power rather than just technology. Its staff have moved between the institute and US regulators.
Build AI tools that strengthen democratic deliberation and collective decision-making (founded by the late Peter Eckersley).
AOI builds open-source tools that let governments hear thousands of citizens at once, used in Taiwan and Tokyo. It is a small nonprofit focused on making AI serve democracy.
Combat bias and harms of AI, especially facial recognition, through research, art and advocacy (founded by Joy Buolamwini).
AJL documents how face recognition and other AI systems fail people of colour and women, and campaigns against their use in airports and policing. Its audits have forced product changes.
Theoretical alignment research (heuristic arguments, low-probability estimation, eliciting latent knowledge) led by Paul Christiano until 2024.
ARC works on the hard mathematical problem of proving that AI systems will behave as intended. Its testing arm became the independent evaluator METR.
Public benefit corporation building Claude with a stated focus on AI safety and a Long-Term Benefit Trust.
Anthropic builds Claude and markets itself as the safety-first lab. Its record is mixed: it held a costly red line against the Pentagon but also softened its own scaling policy and paid a record copyright settlement.
Evaluate frontier models for deceptive and 'scheming' behaviour and build a science of scheming.
Apollo is a London lab that tests whether AI models will lie, sandbag, or try to evade shutdown. Its findings appear in the labs' own model reports.
Studying and shaping the internets impact on society, including AI.
One of the oldest internet-and-society research centers, now running major programs on AI ethics and accountability.
Reduce societal-scale risks from AI through research, field-building, and policy advocacy.
CAIS is a San Francisco nonprofit that does safety research and pushes for AI laws. It wrote the famous one-line 'extinction risk' statement and helped draft California's first frontier-AI bills.
Academic centre led by Stuart Russell developing provably beneficial AI and advocating for AI regulation.
CHAI is Stuart Russell's Berkeley research centre on making AI systems defer to humans. Its leadership has moved from theory into public campaigning for binding limits on AI.
Aligning technology with humanitys best interests, from social media to AI.
The team behind The Social Dilemma and the AI Dilemma talks; the most effective mass-audience communicator on AI race risks.
Data-driven policy analysis on AI and national security for US policymakers.
CSET is a Georgetown think tank that briefs Congress and agencies on AI, chips and China. It is now led by Helen Toner, the former OpenAI board member who voted to remove Sam Altman.
Research and talent pipeline on how humanity can navigate the transition to advanced AI, with a focus on frontier-lab and government policy.
GovAI is a UK think tank studying how governments and labs should manage powerful AI. It trains many of the people who now work inside safety institutes and lab policy teams.
Campaign for binding national and international regulation to prevent superintelligence, via direct outreach to legislators.
ControlAI lobbies lawmakers one by one to treat superintelligence as an extinction-level risk needing binding law. It has persuaded over a hundred UK parliamentarians to sign on.
Steward of open licences; developing norms for how AI may reuse openly shared content.
Creative Commons, which created the licences behind Wikipedia and millions of photos, is trying to set rules for how AI companies should use openly shared work. Its 'signals' are voluntary and unenforced.
Ground technology policy in evidence about real-world social impacts, with emphasis on labour, governance and affected communities.
Data & Society studies what AI actually does to workers, governments and communities, and turns that into policy advice. Its recent work questions whether 'a human in the loop' is meaningful oversight.
Community-rooted, independent AI research free from Big Tech influence, founded by Timnit Gebru after her exit from Google.
DAIR is Timnit Gebru's research institute focused on the harms AI causes today, especially to marginalised communities and data workers. It is deliberately independent of tech company funding.
Open research collective building open models, datasets and evaluation tools.
EleutherAI is a volunteer-born lab that builds open AI models and, in 2025, showed a competitive model can be trained on fully licensed data. Its evaluation tools are industry standard.
Youth-led AI policy advocacy organisation that co-sponsored California's frontier-AI bills.
Encode is a tiny nonprofit started by a teenager that helped write California's first frontier-AI law. OpenAI subpoenaed it during that fight, drawing accusations of intimidation.
Track and forecast the trajectory of AI (compute, data, hardware, benchmarks) with public datasets.
Epoch is the go-to source for numbers on how fast AI is scaling. A 2025 controversy over undisclosed OpenAI funding pushed it to publish a full funding transparency page.
Pioneering research on existential risk, AI safety and the long-term future.
Oxford institute that hosted Nick Bostrom and Toby Ord and seeded much of existential-risk research; closed in 2024.
Steer transformative technology away from extreme large-scale risks through letters, grants, policy work, and public campaigns.
FLI is the organisation behind the best-known AI open letters and an annual report card on AI companies. It also funds grassroots groups and lobbies for regulation in the US and EU.
Alphabet's AI research and frontier-model unit (Gemini), operating under Google's AI Principles and Frontier Safety Framework.
Google DeepMind builds Gemini and publishes one of the more detailed safety frameworks. In 2025 its parent dropped its no-weapons promise and moved into military contracts over staff objections.
Commercial platform hosting open models and datasets, with an active public-policy and ethics team.
Hugging Face is the main hub where open AI models are shared, and it lobbies for open-source-friendly rules. Its energy-rating project exposes that closed-model companies won't disclose their power use.
Policy research on frontier AI security, compute governance and US-China dynamics; runs an AI policy fellowship.
IAPS is a small Washington-oriented research group focused on how to secure advanced AI and the chips it runs on. It states it refuses corporate money to protect its independence.
Originally technical alignment research; since 2024 pivoted to policy advocacy for an international halt to frontier AI development.
MIRI is the oldest AI-risk organisation and once did mathematical alignment research. It now argues publicly that the world should stop building superhuman AI until it can be made safe.
Meta's AI division, historically the leading open-weights frontier lab (Llama), now pursuing 'personal superintelligence'.
Meta championed open-weight AI with Llama but refused Europe's voluntary code and is now moving toward closed models in its race for superintelligence. Independent graders rank its safety practices near the bottom.
Independent pre-deployment evaluation of frontier models for autonomous and dangerous capabilities.
METR is the outside tester the big labs let poke at their models before release. It measures how long and how autonomously AI agents can work, and publishes what it finds.
European frontier lab releasing open-weight and commercial models, positioned as the EU's sovereign AI champion.
Mistral is France's answer to the US labs, releasing many open models. It was first to back the EU's code of practice and first to publish a full environmental accounting of a model.
Promote trustworthy, open and public-interest AI as part of its internet-health mission.
Mozilla, the Firefox nonprofit, argues that open-source AI is safer and fairer than closed systems. Its 2026 report tracks how open models compete with Big Tech.
Build AGI that 'benefits all of humanity' (Charter); now a public benefit corporation controlled by a nonprofit foundation.
OpenAI makes ChatGPT and started as a nonprofit promising to put humanity first. It has since become a for-profit giant, loosened its own safety rules, and used subpoenas against small critics.
Multi-stakeholder nonprofit convening companies, civil society and academia to develop responsible-AI guidance.
PAI brings the big AI companies and civil-society groups into one room to write voluntary best-practice guides. Critics say its corporate funding limits how hard it can push.
Grassroots movement demanding an international pause on training the most powerful AI systems.
PauseAI organises legal, nonviolent street protests asking governments to halt the race to ever-more-powerful AI. It grew from a Dutch founder's initiative into chapters worldwide.
Nonpartisan research for public policy, now including AI security and geopolitics.
The original Cold War think tank, now producing influential work on AI competition, export controls and biosecurity risk.
Research on 'AI control': keeping AI safe even if models are misaligned, plus empirical alignment studies.
Redwood studies how to keep AI systems in check even if they secretly have the wrong goals. Its 2024 work with Anthropic showed a model pretending to go along with training.
Auditing and rating the safety practices of frontier AI companies.
Independent watchdog that publishes ratings of how well AI labs manage risk, so safety claims can be compared.
Advancing AI research, education and policy to improve the human condition.
The academic home of human-centered AI and the annual AI Index, the standard statistical reference on the state of the field.
Civil-disobedience group demanding a permanent ban on AGI development.
Stop AI is a small direct-action group that blocks doors and stages hunger strikes to demand a ban on AGI. It split from PauseAI over tactics and has had turbulent leadership.
Aligning AI through better governance, rules and institutions.
Key civil-society player in European and global AI governance, active in the EU AI Act and summit processes.
Watchdog nonprofit tracking whether AI companies keep their safety promises.
The Midas Project is a small watchdog that keeps score on AI companies' safety pledges and publishes when they slip. Its trackers are the main public record of the Seoul commitments' follow-through.
Government body that tests frontier models pre-deployment and researches AI security risks.
The UK's state AI-testing lab, the first of its kind, examines frontier models before launch. In 2025 it rebranded to 'security' and narrowed its remit away from bias and societal harms.
NIST body serving as industry's point of contact for voluntary AI testing, standards and national-security evaluations.
The US government's AI testing centre, created under Biden and rebranded under Trump. It now tests all major US labs' models voluntarily and benchmarks Chinese models.
Elon Musk's frontier lab building Grok, integrated with the X platform.
xAI builds Grok and has the weakest documented safety follow-through among major labs: missed deadlines, hate-speech and deepfake scandals, and partial participation in industry codes.