Industries / Food, farming and forestry / Tobacco
Regulation and labeling
The science that tied tobacco to disease, and the laws, treaties and court settlements that followed it: health warnings, limits on what packs may say, and rules on how products are sold.
7
Milestones achieved, from 1950 to 2012.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1950
Smoking linked to lung cancer
Two case-control studies, which compare people who have a disease with similar people who do not, asked hospital patients about their smoking. In London, Richard Doll and Austin Bradford Hill interviewed 649 men with lung cancer at 20 hospitals and found that 0.3% had never smoked, against 4.2% of matched men without cancer; they estimated that people smoking 25 or more cigarettes a day had about 50 times the risk of non-smokers, a figure they called admittedly speculative. In the US, Ernst Wynder and Evarts Graham found that only 1.3% of 605 men with the common types of lung cancer were non-smokers, against an estimated 14.6% of other male patients, as Doll and Hill summarize their work.
Credit
Richard Doll and Austin Bradford Hill (British Medical Journal, 30 September 1950); Ernst Wynder and Evarts Graham (Journal of the American Medical Association, May 1950)
1964
The Surgeon General's report
On 11 January 1964 US Surgeon General Luther Terry released the report of a ten-member advisory committee that, with more than 150 consultants, had reviewed more than 7,000 scientific articles. It held smoking responsible for a 70% higher death rate among smokers than non-smokers, and found that average smokers had 9 to 10 times the lung cancer risk of non-smokers, and heavy smokers at least 20 times; it also described the habit as "an habituation rather than an addiction". In 1965 Congress required a health warning on every cigarette pack sold in the US.
Credit
Luther Terry and the Surgeon General's Advisory Committee on Smoking and Health, convened after a report by the Royal College of Physicians of London
1998
Settlements open company files
Minnesota and Blue Cross Blue Shield of Minnesota sued the cigarette makers in 1994 and settled on 8 May 1998; later that year 52 state and territory attorneys general signed the Master Settlement Agreement with the four largest US makers, which pays the states every year for as long as the companies sell cigarettes in the US and bans cartoons in tobacco advertising. Both settlements made the companies' internal records public: about 35 million pages when the Minnesota and UK (Guildford) depositories opened in 1998 and 1999, and about 70 million pages by 2009, by the count of Mayo Clinic researchers.
Credit
the State of Minnesota and Blue Cross Blue Shield of Minnesota; the attorneys general who signed the Master Settlement Agreement
2001
Low-tar numbers shown misleading
For decades US brands were rated by the tar and nicotine that a smoking machine drew from them under a Federal Trade Commission method, and "light" and "low tar" brands scored low largely because of tiny ventilation holes in the filter that let air dilute the smoke. In October 2001 the National Cancer Institute's Monograph 13 concluded that smokers cover the holes with their lips or fingers and take larger or faster puffs, so the machine numbers "do not offer smokers meaningful information", and found no meaningful difference in risk between brands. The 2003 WHO treaty and the US 2009 Tobacco Control Act both ban terms such as "light", "mild" and "low".
LessonA test that measures the product without the person using it can be passed by design changes that do not change what the person receives.
Credit
US National Cancer Institute
2001
Pictures on cigarette packs
Canada became the first country to require picture health warnings on cigarette packs, in 2001, by the count of the Canadian Cancer Society, which tracks pack warnings worldwide. By November 2025, 140 countries and territories, with 66% of the world's population, required picture warnings, and Canada (2024) and Australia (2025) had begun requiring a warning printed on each cigarette.
Credit
Government of Canada; Canadian Cancer Society (tracking)
2005
A global tobacco treaty
The World Health Assembly adopted the WHO Framework Convention on Tobacco Control on 21 May 2003, and it entered into force on 27 February 2005; it has 183 Parties. Article 11 gives each Party three years to require health warnings covering at least 30% of a pack's main display areas, and preferably 50% or more, and to ban terms such as "low tar", "light", "ultra-light" and "mild".
Credit
World Health Organization and its member states
2012
Plain packs upheld at WTO
Australia was the first country to require plain packaging, in 2012: packs with no brand colors, logos or design, in a standard shape and color so that all brands look alike. The Dominican Republic and Honduras, among others, challenged the law at the World Trade Organization; a panel rejected their claims on 28 June 2018, and the Appellate Body upheld the panel on 9 June 2020. By November 2025, 27 countries and territories had adopted plain packaging, by the Canadian Cancer Society's count.
Credit
Government of Australia (Tobacco Plain Packaging Act 2011)
Watch for
A smoke-free generation law
The UK's Tobacco and Vapes Act 2026, which received Royal Assent on 29 April 2026, makes it an offense in England and Wales to sell tobacco, herbal smoking products or cigarette papers to anyone born on or after 1 January 2009, from 1 January 2027, so the age limit rises by one year every year. New Zealand had passed a similar rule and repealed it in February 2024, together with plans for very-low-nicotine cigarettes and far fewer tobacco retailers (Nip and colleagues, New Zealand Medical Journal, 2025).
Achieved whenthe sales rule has been in force for a year and a public body publishes figures on how it was enforced.
Claimed by
UK Parliament, April 2026
AI across industry archives
Daniel Akselrad and Robert Proctor describe INDRA, a research platform that lets a large language model work across the UCSF Industry Documents Library, ToxicDocs and other archives of company records, confined to a set of documents the user picks and tagging which statements come from the documents and which from the model. Their preprint gives three case studies and no measured accuracy.
Achieved whenan outside group tests it on questions with known answers and publishes how often the documents it cites exist and support its statements.
Claimed by
Daniel Akselrad and Robert Proctor (arXiv preprint), September 2026
Open
01
Measure pack warnings from photos
a program that takes photos of a cigarette pack and reports what share of the front and back the health warning covers, whether the warning picture matches the government's published image, and whether the required text is present, then checks each result against that country's law.
The briefCredit · unclaimed02
Find the document behind claims
a public test set of claims from peer-reviewed tobacco-control papers, each paired with the internal company document the paper cites by its archive ID or Bates number (the page number stamped on documents produced in a lawsuit), and a scored comparison of how often keyword search, embedding search and AI research tools find that document and quote it correctly.
The briefCredit · unclaimed
Open 01 · the brief
Measure pack warnings from photos
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
a program that takes photos of a cigarette pack and reports what share of the front and back the health warning covers, whether the warning picture matches the government's published image, and whether the required text is present, then checks each result against that country's law.
Why it is still open
Pack audits are coded by hand. The Johns Hopkins Tobacco Pack Surveillance System (TPackSS) holds photos of more than 8,600 packs from 14 low- and middle-income countries, and its codebook has coders measure each pack face in centimeters. A 2025 study of 191 Mexican packs coded this way found that 85% met the size rule but only 27% showed the warning image exactly as the government specified (Grilo and colleagues). Published computer-vision checks look for warnings in social media posts, such as influencers' Instagram posts for little cigars and cigarillos (Wu and colleagues, 2023), not at a warning's size or accuracy on a pack.
tpackss.globaltobaccocontrol.orgtpackss.globaltobaccocontrol.orgebi.ac.ukebi.ac.uk
Built with what exists
TPackSS pack photos and codebooks; governments' published warning images; OpenCV for finding the pack face and the warning's edges; vision-language models such as Gemini, Claude or GPT for reading warning text; a table of each country's warning rules built from the laws listed in the Canadian Cancer Society's international report.
First step
ask the TPackSS team for permission to use the Mexico photos, then measure 50 packs by hand and with a first script and compare the two.
Done looks like
on 500 packs that TPackSS coders have already measured, warning-area estimates within 2 percentage points of the human measure for at least 90% of packs, and the same pass or fail call on size and image accuracy for at least 95%, with code and results published.
Open 02 · the brief
Find the document behind claims
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
a public test set of claims from peer-reviewed tobacco-control papers, each paired with the internal company document the paper cites by its archive ID or Bates number (the page number stamped on documents produced in a lawsuit), and a scored comparison of how often keyword search, embedding search and AI research tools find that document and quote it correctly.
Why it is still open
The archive is large: the settlements had released about 70 million pages by 2009 (Hurt and colleagues). The TREC Legal Track, run from 2006 to 2009 on a collection drawn from these documents, tested lawyers' discovery requests, not research claims. INDRA (2026) reports case studies without a measured accuracy, and no public test of finding cited tobacco documents was found.
Built with what exists
the UCSF Industry Documents Library and its search; the Europe PMC API to find papers that cite industry documents; the OCR-IDL text annotations of the library (furkanbiten/idl_data on GitHub); sentence-transformers for embedding search; large language models such as Claude for reading and quoting documents (see the Note "Seven more flagged").
First step
pull 20 papers from the journal Tobacco Control that cite UCSF document IDs, write down 60 claims with their cited documents, and time how long a manual search takes to find each one.
Done looks like
300 claims from at least 50 papers, each with its cited document; for at least three search methods, the share of claims whose cited document appears in the top ten results and the share of quotes that match the document word for word; the set published so others can rerun it.