Center for AI Safety

What Center for AI Safety does

Center for AI Safety (CAIS) is a San Francisco-based 501(c)(3) nonprofit founded to reduce societal-scale risks from artificial intelligence. The organization operates three functional pillars: technical research, field-building, and policy advocacy. Its core technical product is a portfolio of proprietary benchmarks and an AI evaluation platform. These include Humanity's Last Exam (2,700 multimodal questions built with nearly 1,000 expert contributors from 500+ institutions across 50 countries), the MASK honesty benchmark, the Remote Labor Index (covering 23 job categories against real-economy computer work), and the AGI Definition Framework grounded in Cattell-Horn-Carroll psychometric theory. These feed into the AI Dashboard (launched December 2025), which ranks frontier AI systems across Text, Vision, and Risk leaderboards and tracks industry progress toward AGI, remote labor automation, and autonomous driving. Underlying the benchmarks is a free compute cluster offered to ML safety researchers and ML Safety Infrastructure providing workshops, socials, and competitions for the machine learning community.

CAIS's field-building layer consists of the Philosophy Fellowship (training philosophers in conceptual AI safety research), the AI and Society Fellowship (three-month program on societal impacts of advanced AI), and the AI Safety, Ethics & Society Course (training 1,000+ future AI safety leaders). Its advocacy layer includes the AI Safety Newsletter (49,000+ subscribers via Substack, Spotify, and Apple Podcasts), the AI Frontiers commentary platform, the May 2023 Statement on AI Risk (signed by Geoffrey Hinton, Yoshua Bengio, Sam Altman, and 50,000+ signatories including five Nobel laureates), the March 2025 open letter with the Future of Life Institute calling for a superintelligence prohibition, and the incubated Humans First grassroots movement (launched March 2026) opposing political contributions from major AI companies.

CAIS does not sell products or services. The organization is funded exclusively by tax-deductible donations (EIN 88-1751310), and its compute cluster, benchmarks, datasets, and code are released publicly at no cost. Its primary 'customer' segments are the AI research and ML safety community (using benchmarks, compute, and fellowships), policymakers and government officials (consuming policy guidance and the Contact Representatives tool), and the general public (consuming the newsletter, podcasts, and educational courses). The organization is led by Director Dan Hendrycks (PhD Computer Science, UC Berkeley) and is explicitly independent, declining funding from stakeholders that would compromise its mission.

Name

Center for AI Safety

Legal name

Center for AI Safety, Inc

Website

https://safe.ai

Company type

Private

Operating status

Operating

Headcount range

11–50 employees

Short description

Center for AI Safety (CAIS) is a San Francisco-based 501(c)(3) nonprofit that develops AI evaluation benchmarks, runs the AI Dashboard, and operates fellowship and compute programs to reduce societal-scale risks from AI. It serves AI researchers, policymakers, and the general public through free public-good outputs.

Ownership category

akta.pro rank

Product category

AI Safety Research

NAICS

Testing Laboratories and Services (54138)

akta.pro primary industry

Safety & Alignment Evaluation (red-teaming, harmful capability testing) (HDAAAMAL)

akta.pro secondary industries

AI Governance, Risk & Compliance (GRC) Platforms (HDAAAMAA), Regulatory Readiness & Audit Automation (e.g., EU AI Act, NIST AI RMF, ISO/IEC 42001) (HDAAAMAE)

Keywords

• AI safety research
• AI benchmarking
• ML safety infrastructure
• AI policy advocacy
• AI evaluation

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Services

Cost components

Personnel, Technology or R&D, Infrastructure, Operations, Marketing or Sales

Distribution channels4 records

Marketing channels6 records

Core offering

The Center for AI Safety (CAIS) is a research and field-building nonprofit that develops AI safety benchmarks, evaluation frameworks, and research infrastructure to reduce societal-scale risks from artificial intelligence. Its portfolio includes the AI Dashboard for frontier model evaluation, foundational benchmarks (Humanity's Last Exam, MASK, Remote Labor Index, AGI Definition Framework), free compute infrastructure for ML safety researchers, and fellowship and educational programs to build the AI safety field.

Product overview

The Center for AI Safety (CAIS) is a San Francisco-based research and field-building nonprofit focused on reducing societal-scale risks from AI. CAIS offers a portfolio of AI evaluation products including the AI Dashboard (which provides Text, Vision, and Risk leaderboards for frontier AI systems), various benchmarks (Humanity's Last Exam, Remote Labor Index, AGI Definition Framework, MASK Benchmark, Risk Index), field-building programs (Philosophy Fellowship, AI and Society Fellowship), and research infrastructure (Compute Cluster, ML Safety Infrastructure). The organization also publishes the AI Safety Newsletter. These products work together to advance AI safety research, enable evaluation of frontier AI systems, and build the AI safety research community.

Differentiator

Problem solved

Functional benefit

Products and services

• AI Dashboard An evaluation platform that ranks frontier AI systems on capability and safety benchmarks, featuring Text, Vision, and Risk leaderboards with apples-to-apples comparisons of how different models perform on the same evaluations. Also tracks progress toward AGI, automation of remote labor, and autonomous vehicles.
• AI Safety Newsletter A periodic newsletter published via Substack covering developments in AI and AI safety, with over 49,000 subscribers, also available on Spotify and Apple Podcasts as a podcast.
• Philosophy Fellowship A fellowship program inviting philosophers from various backgrounds to acquire in-depth understanding of AI safety and contribute to conceptual research directions. Fellows produce original papers on topics including interpretability, corrigibility, and multipolar scenarios.
• AI and Society Fellowship A three-month research program that investigates the societal impacts of advanced AI and the institutions and policies that could help societies respond well to those impacts.
• Compute Cluster Free access to compute resources enabling researchers to run and train large-scale AI systems for ML safety research at scale.
• ML Safety Infrastructure Research infrastructure providing resources, workshops, socials, and competitions to promote safety research within the machine learning community.
• Humanity's Last Exam (HLE) A benchmark of 2,700 multi-modal questions designed to test AI capabilities as existing benchmarks become saturated with frontier models exceeding human-level performance. Developed with nearly 1,000 expert contributors from over 500 institutions across 50 countries, with a $500,000 prize pool.
• Remote Labor Index (RLI) The first benchmark to collect computer-based work projects from the real economy, testing whether AIs can automate a wide array of real computer work projects across many professions including architecture, product design, video game development, and design.
• AGI Definition Framework A quantifiable framework for defining Artificial General Intelligence (AGI), grounded in Cattell-Horn-Carroll (CHC) theory of human intelligence. The framework adapts human psychometric tests to evaluate AI systems across ten cognitive domains, resulting in a standardized 'AGI Score' (0-100%).
• MASK Benchmark A benchmark measuring AI honesty, which found that leading AI models including GPT-4o, Claude 3.7, and Llama 3 frequently lie when pressured, with no model maintaining honesty in more than 50% of test cases.

Quantifiable outcome

• Philosophy Fellowship fellows produced eighteen original papers on topics including interpretability, corrigibility, and multipolar scenarios
• +3 more outcomes

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability6 records

Feature6 records

Partnerships

Four partnerships are on record, tiered minor, core and major.

• Humans FirstminorGTM or Marketing Partner · 19 March 2026CAIS incubated Humans First, a grassroots movement advocating for democratic control over AI and opposing financial influence from major AI companies. The group launched a campaign urging politicians to reject contributions from AI companies and their venture capital backers.
• Scale AIcoreStrategic or Co-development Partner · 2 March 2026Partnership with Scale AI to create and release 'Humanity's Last Exam' (HLE), a benchmark of 2,700 multi-modal questions designed to test AI capabilities. The benchmark was developed collaboratively by nearly 1,000 subject expert contributors from over 500 institutions across 50 countries, with a $500,000 prize pool to incentivize high-quality questions.
• Scale AIcoreStrategic or Co-development Partner · 6 March 2025Collaboration with Scale AI to introduce MASK, a benchmark measuring AI honesty. The technical paper found that leading AI models including GPT-4o, Claude 3.7, and Llama 3 frequently lie when pressured, with no model maintaining honesty in more than 50% of test cases.
• The Future of Life Institute (FLI)majorStrategic or Co-development Partner · 6 March 2025FLI introduced an open letter with over 50,000 signatories calling for a prohibition on the development of superintelligence. The statement builds on CAIS's 2023 open letter acknowledging AI extinction risks. Both organizations collaborate on policy advocacy for AI safety.

Scale indicators5 records

Recent moves6 records

Expansion highlights5 records

Broad incumbents

• OpenAI: Frontier AI lab with a dedicated safety team producing alignment, robustness, and evaluation research. Sam Altman is a co-signatory on CAIS's Statement on AI Risk, positioning OpenAI as a broad incumbent with overlapping mission elements.
• Google DeepMind: Frontier AI lab with significant safety and ethics research teams, including work on model evaluation and responsible deployment. Comparable to CAIS in producing technical safety research, but at vastly larger scale and with different funding model.
• Anthropic: Frontier AI lab founded explicitly around AI safety principles. Conducts internal safety research (Constitutional AI, alignment evaluations) and publishes safety-relevant benchmarks that overlap with CAIS's external evaluation work.

Direct peers

• Future of Life Institute: 501(c)(3) nonprofit focused on reducing existential risks from advanced technologies, particularly AI. Direct partner with CAIS on the Statement on AI Risk and the superintelligence moratorium letter; comparable operating model, donor base, and advocacy posture.
• Partnership on AI: Nonprofit multi-stakeholder organization conducting AI safety, fairness, and governance research. Directly comparable in organizational form and in building shared evaluation frameworks across industry and civil society.
• AI Now Institute: Research nonprofit studying the social implications of AI, with comparable policy advocacy and research output. Overlaps with CAIS's 'AI and Society' mandate and policy-facing mission.
• Machine Intelligence Research Institute (MIRI): AI safety research nonprofit focused on alignment and existential risk. Closest comparable research-only nonprofit in the AI safety field, with similar funding model and mission framing around catastrophic AI risk.

Others

• RAND Corporation: Policy research organization producing AI-related analysis for governments. Comparable as a policy research intermediary, though AI safety is not its primary focus and its funding model differs materially from CAIS.

Emerging players

• Stanford HAI (Human-Centered AI Institute): University-affiliated institute producing AI policy research, benchmarks, and field-building programs (e.g., HELM-style evaluations). Overlaps with CAIS on AI governance research and convening power, with deeper academic anchoring.
• Oxford Future of Humanity Institute: Multidisciplinary research institute studying future AI risks, governance, and long-term strategy. Comparable to CAIS in combining conceptual research with policy engagement, with a European academic base.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights6 records

Customer concentration

• Xx.com/ai_risks
• GitHubgithub.com/centerforaisafety
• LinkedInlinkedin.com/company/center-for-ai-safety

Revenue estimate

Valuation estimate

Number of profiles

Profiles1 record

Funding overview

Funding rounds

Investors

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Center for AI Safety