Canada, 2017 — Budget Implementation Act
An AI law that is really a line in a budget: $125 million to a research institute.
Legislatures learned to talk about AI. They have not learned to pass laws about it.
Chapter 5 of the AI Index 2022 reads the legislative record of 25 countries, every US state, and the US Congress for the year 2021. Mentions of AI are multiplying everywhere. Bills are multiplying almost as fast. Laws are not — 130 AI-related bills were proposed in Congress and three passed. The second half of the chapter follows the money instead, and finds that American public AI spending is, overwhelmingly, defense spending. The numbers:
18AI-related bills passed into law across 25 countries in 2021 (1 in 2016)
2% of US federal AI-related bills that became law in 2021 (130 proposed, 3 passed)
131AI-related bills proposed in US state legislatures in 2021 (2 in 2012); 26 passed
1,323AI mentions in the legislative proceedings of 25 countries in 2021, 7.7× the 2016 level
9.26US$ billions the US Department of Defense allocated to AI R&D in FY2021, across 500 programs
1.79US$ billions of US federal AI-related contract spending in 2021 (2020 peak: 1.97)
5.1 — Six years, 25 countries, 55 laws
The AI Index searched the legislative records of 25 countries for bills containing the words “artificial intelligence,” in various languages, from 2016 to 2021. The entire six-year harvest is 55 laws — and close to a third of them arrived in the final year.
The annual count rose from just 1 law in 2016 to 18 in 2021. No country dominates. Spain, the United Kingdom, and the United States each passed three in 2021 — the joint highest — followed by Belgium and Russia with two each, and France, Germany, Italy, Japan, and South Korea with one each. The 2021 chart lists ten countries in total, which means the other 15 in the sample passed nothing that year.
The cumulative picture is thinner than the headlines
Summed across 2016–21, the United States leads with 13 laws, its first coming in 2017. Russia follows with six; Belgium, Spain, and the United Kingdom have five each; France, Italy, and South Korea four each; Japan three; China two; and Brazil, Canada, Germany, and India one apiece. That is the whole list. Fourteen countries out of 25 have passed at least one AI-related law in six years.
Two caveats worth carrying forward
- The count includes only laws passed by national legislative bodies — congresses and parliaments — with “artificial intelligence” in the title or body of the bill text. Regulations, executive orders, and subnational law fall outside it entirely.
- The 25 countries are Australia, Belgium, Brazil, Canada, China, Denmark, Finland, France, Germany, India, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Russia, Singapore, South Africa, South Korea, Spain, Sweden, Switzerland, the United Kingdom, and the United States.
- Passing a law is not the same as regulating AI. Several of the 55 are appropriations acts that happen to fund an AI program — Canada’s Budget Implementation Act 2017, which authorized a $125 million payment to the Canadian Institute for Advanced Research for a pan-Canadian AI strategy, is one of them.
What an “AI law” actually says
Seven of the 55, singled out by the AI Index and spanning 2017 to 2021. Read together they show how little the category holds in common: a research grant, a health-data mandate, a regulatory sandbox, an exam-marking pilot, a deepfake research program, an ethics committee, and an environmental monitor.
China, 2019 — Basic Medical and Health Care Law
AI written into health policy: promote its use, and build the data standards it needs.
Russia, 2020 — Federal Law 123-FZ
A five-year AI regulatory experiment confined to one city, starting July 2020.
United Kingdom, 2020 — Supply and Appropriation Act
Authorizing the exams regulator to explore AI for marking high-stakes qualifications.
United States, 2020 — IOGAN Act
Directing the National Science Foundation to fund research into deepfake outputs.
Belgium, 2021 — Decree on support for job seekers
If AI tools are used in digitization, an Ethics Committee must advise first.
France, 2021 — Law 2021-1485
A monitoring system for the environmental impact of emerging digital technology, AI named explicitly.
5.1 — In Washington, 2%. In the state capitols, 20%.
The distance between proposing an AI bill and passing one is the cleanest number in the chapter. It is ten times wider at the federal level than at the state level, and the states are where nearly all the actual lawmaking happens.
In 2015 exactly one AI-related bill was proposed in the US Congress. In 2021 there were 130. Three passed. That is 2% of the year’s federal AI bills becoming law, and it describes the whole period: proposals climb steeply, enactments barely move. The gap was at its widest in 2021, the year the proposals peaked.
The states pass a fifth of what they propose
Bloomberg Government has tracked state-level bills containing AI keywords — artificial intelligence, machine learning, algorithmic bias — since 2012. That year there were two, both from New Jersey assembly member Annette Quijano, directing the state Motor Vehicle Commission to establish driver’s license endorsements for autonomous vehicles. In 2021 there were 131, of which 26 became law: 20%, or one in five, against one in fifty federally.
Where the bills come from
- As of 2021, 41 of the 50 states had proposed at least one AI-related bill. Massachusetts leads the decade with 40 since 2012, ahead of Hawaii (35), New Jersey (32), New York (31), California (29), and Illinois (28).
- In 2021 alone the order changes completely: Massachusetts 20, Illinois 15, Alabama 12. Volume in any single year says more about one legislature’s session than about a state’s long-run posture.
- AI lawmaking has a partisan slope. In 2018 Democratic state legislators sponsored just two more AI bills than Republicans; in 2021 they sponsored 39 more — 79 against 40.
5.1 — Talk is the fastest-growing indicator
If laws are the lagging measure of political interest in AI, mentions are the leading one. Every mention series in the chapter grows faster than the legislation that accompanies it — and so does the output of the organizations advising the legislators.
Bloomberg Government counts AI keyword mentions in US congressional records, split across legislation, committee reports, and Congressional Research Service reports. The 116th session (2019–20) logged 506 mentions — nearly 3.4 times the 115th session’s 149 (2017–18), and about 30 times the 114th session’s 17 (2015–16). The 117th had already recorded 295 by the end of 2021, halfway through its term, putting it on course for the highest count since tracking began in 2001.
Parliaments elsewhere are louder still
Across the same 25 countries, AI mentions in legislative proceedings reached 1,323 in 2021 — 7.7 times the 2016 level. Spain led the year with 269, ahead of the United Kingdom (185), the United States (132), Australia (122), and Japan (95). Over the full six years the ordering is different: the United Kingdom dominates with 939 mentions, followed by Spain (559), Japan (466), the United States (422), Australia (410), and Singapore (282).
And the people advising them
- The AI Index tracks 55 US-based organizations that published AI policy papers over the past four years: 19 think tanks and policy institutes, 14 university institutes and research programs, nine civil society organizations and consortiums, nine industry and consultancy organizations, and four government agencies.
- Their combined output tripled between 2018 and 2020, peaking at 273 papers, then eased slightly to 210 in 2021.
- The 2021 agenda was led by privacy, safety, and security (62 papers as a primary topic), innovation and technology (62), and ethics (59). Democracy was the primary topic of only two papers, but a secondary topic of 29. Energy and the environment, the humanities, the physical sciences, and the social and behavioral sciences drew almost no attention at all.
5.2 — Where American public AI money actually goes
The chapter’s second half narrows to one country: the United States, one of the world’s largest public-sector investors in AI. Six questions, and a running theme — almost every dollar has a defense label on it.
How much does the US spend on nondefense AI R&D?
In fiscal year 2021, nondefense federal agencies allocated $1.53 billion to AI R&D — roughly 2.7 times the $0.56 billion of FY2018, with $1.11 billion in FY2019 and $1.43 billion in FY2020 in between. The FY2022 request rises 8.8% to $1.67 billion. The figures come from a December 2021 National Science and Technology Council report covering agencies in the NITRD program and the National AI Initiative, and they exclude classified AI R&D by the defense and intelligence agencies.
And with defense included?
The scale changes entirely. For FY2021 the Department of Defense allocated $9.26 billion across 500 AI-related R&D programs, 6.68% above the $8.68 billion of FY2020, and it has requested $10.00 billion for FY2022 — a figure likely to grow once further requests and congressional appropriations are counted. Within the department the Navy was the top spender in FY2021 and asks for the most again in FY2022 ($1.86 billion), ahead of the Army ($1.77 billion), the Office of the Secretary of Defense ($1.1 billion), and the Air Force ($883 million).
Can that $10 billion be trusted?
The chapter flags it as a caveat rather than a finding. Bloomberg Government’s keyword search of DOD budget lines produces $10.0 billion for FY2022 AI-specific R&D; the DOD’s own measurement of the same year produces $874 million. The discrepancy is definitional: a research project that uses AI for cyber defense may count its people, hardware, and operations costs inside an AI-related budget request even though the AI software component is much smaller.
What about contracts, rather than budget requests?
In 2021 federal departments and agencies spent $1.79 billion on AI-related contracts — nearly double the roughly $920 million of 2018, but a slight decrease from the 2020 peak of $1.97 billion. The dip arrived while every legislative indicator in the chapter was still climbing. Bloomberg Government classifies a contract as AI-related by adding up transactions whose titles or descriptions contain any of more than 100 AI-specific keywords, so some projects counted here have a relatively small AI component.
Which departments get the contract money?
One, mostly. The Department of Defense spent $1.14 billion on AI contracts in 2021, roughly five times the next department, Health and Human Services ($234 million), followed by NASA ($159 million), Homeland Security ($81 million), and Commerce ($49 million). Cumulatively from 2000 to 2021 the DOD has spent $5.20 billion, against NASA’s $1.41 billion, HHS’s $0.70 billion, Homeland Security’s $0.45 billion, and the Treasury’s $0.32 billion.
What do those contracts actually buy?
The largest AI-related contract signed in 2021 by each of the five top-spending departments: $70 million (DOD) for prototypes in automotive cybersecurity, vehicle safety technologies, autonomous vehicles and intelligent systems; $37 million (Homeland Security) for towers capable of autonomous surveillance; $20 million (HHS, through BARDA) for optical imaging devices and machine learning algorithms to help classify and heal wounds and burns; $14 million (NASA) for lunar robots able to navigate the moon’s south pole; and $13 million (Commerce) for an AI prototype to improve patent search.
The chapter in five lines
Headline findings from Chapter 5 · AI Policy and Governance, reporting on 2021.
An AI Index analysis of legislative records on AI in 25 countries shows that the number of bills containing “artificial intelligence” that were passed into law grew from just 1 in 2016 to 18 in 2021.
The federal legislative record in the United States shows a sharp increase in the total number of proposed bills that relate to AI from 2015 to 2021, while the number of bills passed remains low, with only 2% ultimately becoming law.
In 2021, of the 131 proposed state bills, 26 were passed into law — 1 out of 5 proposed bills, significantly higher than the federal level, where 1 out of every 50 proposed bills became law.
The current congressional session, the 117th, is on track to record the greatest number of AI-related mentions since 2001, with 295 mentions by the end of 2021, halfway through the session, compared to 506 in the previous, 116th session.
Bloomberg Government’s analysis shows the department is requesting $10.0 billion for AI-specific R&D in FY 2022. However, DOD’s own measurement produces a smaller number of $874 million.
Read Chapter 5 in full
Chapter 5 (sections 5.1–5.2) with every figure, table, and methodological note is free from Stanford HAI. Or head back to the report highlights and the full chapter overview.
Open the AI Index 2022 →