Skip to content
Longterm Wiki

Center for AI Safety (CAIS)

Safety Organization
Founded 2022 (4 years old)HQ: San Franciscosafe.ai

Also known as: CAIS

Structured Facts
Database Records
Revenue
$10M
as of 2024
Total Funding Raised
$33M
as of 2025
Founded Date
2022

Key People

7
VK
Varun Krovi
Executive Director, CAIS Action Fund; Director of Government Relations & Public Policy
Leads CAIS Action Fund and government relations. 15+ years policy/advocacy experience, former Capitol Hill Chief of Staff. Confirmed via LegiStorm and LinkedIn as of 2026-04-22.
OZ
Co-Founder and Managing Director
2022 – present
Co-founded CAIS in 2022 with Dan Hendrycks. Confirmed on Wikipedia and The Org as of 2026-04-22.
AZ
Andy ZouFounder
Co-Founder
2022 – present
Co-founded CAIS; also PhD student at CMU and founder of Gray Swan AI. Confirmed on personal site and Future of Life Institute as of 2026-04-22.
AZ
Research Scientist
Lead author of adversarial attacks and representation engineering papers
TW
Policy Director
SW
Key Legislator (SB 1047 sponsor)
Feb 2024 – present
California State Senator. Introduced SB 1047 February 2024. Passed legislature August 2024, vetoed September 2024. Also authored SB 53 (2025). Per Wikipedia.
JE
Chief Operating Officer
Start date unknown; Confirmed on safe.ai/about as of 2026-03-16.

Funding History

6
Open Philanthropy General Support 2024grant2024
$8.5MLed by Open Philanthropy
$8,500,000 general support grant from Open Philanthropy in 2024. Confirmed via OP 2024 progress report.
openphilanthropy.org
SFF General Support 2024 (Jaan Tallinn)grant2024
$1.1MLed by Jaan Tallinn
$1,146,000 from Jaan Tallinn via Survival and Flourishing Fund, 2024.
survivalandflourishing.fund
Open Philanthropy General Support 2023grantApr 2023
$4MLed by Open Philanthropy
$4,000,000 general support grant from Open Philanthropy, April 2023.
openphilanthropy.org
SFF General Support 2023-H2 (Jaan Tallinn)grant2023
$909KLed by Jaan Tallinn
$909,000 from Jaan Tallinn via Survival and Flourishing Fund, H2 2023.
survivalandflourishing.fund
SFF General Support 2023-H1 (Jaan Tallinn)grant2023
$22KLed by Jaan Tallinn
$22,000 from Jaan Tallinn via Survival and Flourishing Fund, H1 2023.
survivalandflourishing.fund
FTX Future Fund Grant 2022grant2022
$6.5MLed by FTX Future Fund
$6.5M received in 2022. FTX bankruptcy estate issued subpoenas to CAIS in October 2023 seeking return of funds. Aligns with CAIS 2022 total revenue of $6.66M (IRS Form 990).
bloomberg.com

All Facts

Financial

Annual Expenses$7.2M20243 pts
As OfValueLink
2024$7.2Mview →
2023$8.1Mview →
2022$817Kview →
Grant Received$1.1M20254 pts
As OfValueLink
2025$1.1Mview →
2024$2.8Mview →
2023$5.5Mview →
2022$5.2Mview →
Net Assets$12M20243 pts
As OfValueLink
2024$12Mview →
2023$8.5Mview →
2022$5.8Mview →
Revenue$10M20243 pts
As OfValueLink
2024$10Mview →
2023$16Mview →
2022$6.7Mview →
Total Funding Raised$33M2025view →

Organization

Founded Date2022view →
HeadquartersSan Franciscoview →

General

Websitehttps://www.safe.ai/view →

Other

board-memberJaan Tallinn2024view →
campaignStatement on AI Risk (May 2023): one-sentence statement 'Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.' Signed by 350+ AI leaders including Geoffery Hinton, Demis Hassabis, Sam Altman, and Dario Amodei.May 2023view →
compensationDan Hendrycks takes $1 annual salary as Executive Director2025view →
infrastructureCompute cluster with 256 NVIDIA A100 GPUs available for AI safety researchers2024view →
key-personDan Hendrycks2025view →
programML Safety Scholars — educational program training hundreds of students in AI safety fundamentals. Includes online course, reading groups, and mentorship.2024view →
publicationThe WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning — benchmark for evaluating dual-use AI capabilities in biosecurity, cybersecurity, and chemical weaponsMar 20243 pts
As OfValueLink
Mar 2024The WMDP Benchmark: Measuring and Reducing Malicious Use With Unlearning — benchmark for evaluating dual-use AI capabilities in biosecurity, cybersecurity, and chemical weaponsview →
Oct 2023Representation Engineering: A Top-Down Approach to AI Transparency — proposes methods to read and control LLM internal representations for safetyview →
Jan 2021Measuring Massive Multitask Language Understanding (MMLU) — widely-used benchmark for evaluating LLM capabilities across 57 academic subjectsview →

Divisions

7
NameDivisionTypeStatusSourceNotesLeadSlugStartDateWebsiteSource check
Field-Buildingprogram-areaactivesafe.aiPrograms to grow the AI safety research community, including the Statement on AI Risk signed by hundreds of researchers and the ML Safety course.
Compute Clusterprogram-areaactivesafe.aiProvides free compute access to academic AI safety researchers. One of the largest non-industry compute resources available for safety research.
Researchteamactivesafe.aiTechnical AI safety research on robustness, interpretability, and alignment. Led by Dan Hendrycks.
Researchteamactivesafe.aiTechnical AI safety research on robustness, interpretability, and alignment. Led by Dan Hendrycks (Executive & Research Director). Confirmed via CAIS about page as of 2026-04-23.Dan Hendrycks
AI and Society Fellowshipprogram-areaactivesafe.ai3-month SF program. $25K stipend. PhD/JD researchers.cais-fellowship
CAIS Compute Clusterlabactivesafe.ai80 A100 GPUs. 150+ researchers. ~100 safety papers, 16,000+ citations. Free access. Schmidt Sciences partnership.cais-compute
CAIS Action Fundprogram-areaactiveaction.safe.ai501(c)(4) advocacy arm. DC-based. Co-sponsored SB 1047. Lobbying ~$490K/yr.Varun KroviCenter for AI Safety Action Fund2023-07action.safe.ai

Entity Events

6
TitleDateEventTypeDescriptionSignificanceSourceSource check
Reported revenue of $10.2M (FY2024)2024milestoneCumulative funding reaches ~$33M since founding ($6.7M in 2022, $16.1M in 2023, $10.2M in 2024).moderateprojects.propublica.org
Statement on AI Risk released2023-05milestoneOne-sentence statement on AI extinction risk attracted signatures from over 350 AI researchers and industry figures, including Turing Award recipients (Hinton, Bengio, Russell) and CEOs of major AI labs (Altman, Amodei, Hassabis).major
MACHIAVELLI benchmark released2023publicationBenchmark for evaluating goal-directed and deceptive behavior in AI systems.moderate
Representation Engineering paper published2023publicationMethods for reading and steering model internal representations.major
"Unsolved Problems in ML Safety" published2022publicationTaxonomy of open technical challenges in machine learning safety, intended partly as a research agenda for the field.major
Founded by Dan Hendrycks and Oliver Zhang2022foundingNonprofit research organization (EIN 88-1751310) focused on technical AI safety research, field-building, and public communication.major

Publications

12
TitlePublicationTypeAuthorsUrlPublishedDateIsFlagshipSourceNotesSource check
Humanity's Last ExampaperLong Phan, Alice Gatti, Ziwen Han, Nathaniel Li et al.arxiv.org2025-01arxiv.org
Introduction to AI Safety, Ethics, and SocietybookDan Hendrycksaisafetybook.com2024-06aisafetybook.comPublished by Routledge. Open-access online + audiobook.
The WMDP Benchmark: Measuring and Reducing Malicious Use With UnlearningpaperNathaniel Li, Alexander Pan, Anjali Gopal et al.wmdp.ai2024wmdp.aiICML 2024. Biosecurity/cybersecurity knowledge unlearning.
Superintelligence StrategyreportDan Hendrycks, Eric Schmidt, Alexandr Wangnationalsecurity.ai2024nationalsecurity.aiCo-authored with former Google CEO and Scale AI CEO
Improving Alignment and Robustness with Circuit BreakerspaperAndy Zou, Long Phan, Justin Wang et al.arxiv.org2024arxiv.orgICML 2024
HarmBench: A Standardized Evaluation Framework for Automated Red TeamingpaperMantas Mazeika, Long Phan, Xuwang Yin et al.harmbench.org2024harmbench.orgICML 2024
Representation Engineering: A Top-Down Approach to AI TransparencypaperAndy Zou, Long Phan, Sarah Chen et al.arxiv.org2023-10arxiv.org
An Overview of Catastrophic AI RiskspaperDan Hendrycks, Mantas Mazeika, Thomas Woodsidearxiv.org2023-06arxiv.org
Statement on AI Riskpolicy-briefCAISaistatement.com2023-05aistatement.comOne-sentence statement signed by Hinton, Bengio, Altman, Amodei, Hassabis
Universal and Transferable Adversarial Attacks on Aligned Language ModelspaperAndy Zou, Zifan Wang, Nicholas Carlini et al.llm-attacks.org2023llm-attacks.orgHighly influential jailbreaking paper
Unsolved Problems in ML SafetypaperDan Hendrycks, Nicholas Carlini, John Schulman, Jacob Steinhardtarxiv.org2021-09arxiv.orgDefines 4 core challenges: robustness, monitoring, alignment, systemic safety
Measuring Massive Multitask Language Understanding (MMLU)paperDan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, Jacob Steinhardtarxiv.org2020-09arxiv.orgMost widely used AI capability benchmark. ICLR 2021.
Internal Metadata
ID: sid_y4bieqSeag
Stable ID: sid_y4bieqSeag
Wiki ID: E47
Type: organization
YAML Source: packages/factbase/data/fb-entities/cais.yaml
Facts: 26 structured (27 total)
Records: 38 in 5 collections