Industries / Health / Healthcare / Workplace health and drug testing
Rules
The laws and court rulings that decide which workers an employer may test, when, and what it may do with a positive result.
7
Milestones achieved, from 1986 to 2019.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1986
Federal workers tested
On 15 September 1986 President Ronald Reagan signed Executive Order 12564, which told US federal employees to refrain from illegal drugs and told every agency head to set up a program to test employees in sensitive positions. Agencies had to give employees 60 days' notice and tell them where to get counselling, and HHS was given the job of setting the scientific and technical guidelines that every agency's testing must follow.
LessonThe order made HHS the scientific authority for workplace testing, and HHS's proposed guidelines of October 2026 still cite it.
Credit
President Ronald Reagan
1989
Testing without suspicion upheld
On 21 March 1989 the US Supreme Court decided two cases. In Skinner v. Railway Labor Executives' Association it upheld federal rules requiring blood and urine tests of railroad crews after major accidents, and allowing breath or urine tests after some safety-rule violations. In National Treasury Employees Union v. Von Raab it upheld urine tests for Customs Service employees seeking jobs in drug interdiction or jobs that require carrying a gun. Taking and testing the samples are searches, the Court held, but safety and the nature of the jobs made them reasonable without a warrant or suspicion of any one worker, and it rejected the lower court's view that tests must be limited to detecting current impairment.
LessonThe exception has limits: on 15 April 1997, in Chandler v. Miller, the Court struck down Georgia's drug test for candidates for state office, calling the need "symbolic, not 'special'".
Credit
US Supreme Court, opinions by Justice Anthony Kennedy
2000
Urine cannot show impairment
On 21 July 2000 the Court of Appeal for Ontario ruled on Imperial Oil's testing policy under the province's human rights code. It held that random breath alcohol tests of employees in safety-sensitive jobs can be justified, because breath tests do show impairment, provided the penalty for a positive result fits the employee's circumstances. It found that the human rights board had gone beyond the complaint in examining drug testing, but since the lower bodies had ruled on it, the court did too: random urine drug tests could not be justified, because a drug test "cannot measure present impairment", only past use.
LessonThe court separated tests of impairment from tests of past use. Its reach is limited: in 2013 the Supreme Court of Canada said that human rights cases from non-union workplaces, such as this one, are "of little conceptual assistance" where a collective agreement governs.
Credit
Court of Appeal for Ontario, reasons by Justice John Laskin
2012
Saliva preferred over urine
In 2012 Fair Work Australia, the national workplace tribunal, found that Endeavour Energy's plan to bring in urine testing would be unjust and unreasonable when oral fluid testing to AS 4760 was available. Urine may miss cannabis smoked in the previous four hours, the time that matters most for impairment, and can show use from several days before. A full bench, a panel of three members, upheld the decision. On 19 August 2015 another full bench allowed Port Kembla Coal Terminal to test at random with both urine and oral fluid. It found that neither test establishes impairment, that workers not knowing which test they will face adds to deterrence, and that most coal export terminals already used urine.
LessonThe tribunal chose between specimens case by case, by which detection window fit the risk, and reached different answers three years apart.
Credit
Fair Work Australia and its successor, the Fair Work Commission
2013
Random tests need a problem
On 14 June 2013 the Supreme Court of Canada, by six votes to three, upheld an arbitration board that had struck down random breath tests of 10% of safety-sensitive employees a year at Irving's paper mill in Saint John, New Brunswick. Arbitrators had allowed testing a worker when there was reasonable cause to think the worker impaired, after an accident or on return from treatment, but had overwhelmingly rejected random testing imposed without a union's agreement unless there was a general problem of substance use at the workplace. The board found that eight alcohol-related incidents in 15 years did not show such a problem. In August 2026 an Ontario arbitrator, Laura Trachuk, struck down the Toronto Transit Commission's random program, which since 2017 had tested about 87% of its 16,000 workers with oral fluid for drugs and breath for alcohol, while accepting oral fluid tests after incidents and for reasonable cause, by the law firm Hicks Morley's account.
LessonIn a unionized Canadian workplace, a dangerous site alone does not justify random testing.
Credit
Supreme Court of Canada, majority reasons by Justice Rosalie Abella
2016
France lets supervisors test
On 5 December 2016 France's highest administrative court, the Conseil d'État, ruled that a quick saliva drug test only reveals recent use, is not a medical laboratory test and needs no doctor, so a supervisor may carry it out if the company's internal rules, the règlement intérieur, provide for it. Random tests were lawful because the rules limited them to "hypersensitive" posts, where being under the influence is a particular danger to the worker and others, bound the employer to secrecy about results, and gave a worker who tested positive the right to a medical counter-test paid for by the employer.
LessonLike the Ontario court in 2000, it accepted that the test shows use, not impairment, but unlike it, allowed random tests for the most dangerous posts.
Credit
Conseil d'État, in a case brought by the builder SOGEA Sud
2019
Cannabis no bar to hiring
A Nevada law of 2019 made it unlawful to refuse to hire someone because a pre-employment test shows marijuana, except for firefighters, emergency medical technicians, drivers who must be tested under federal or state law and jobs the employer judges could affect others' safety. A new hire tested in the first 30 days may pay for a second test to rebut the first. From 1 January 2024 California bars employers from penalizing a person for using cannabis off the job and away from work, or for a test that finds non-psychoactive cannabis metabolites, the breakdown products that linger after the effects wear off. It still allows tests that do not screen for those metabolites and still bans use or impairment at work.
LessonCalifornia ties a lawful test to what it detects, the line the Ontario court drew in 2000.
Credit
Nevada Legislature; California Legislature
Watch for
All marijuana to Schedule III
A final order effective on 28 April 2026 moved FDA-approved drugs containing marijuana, and marijuana covered by a state medical marijuana licence, from Schedule I to Schedule III of the US Controlled Substances Act. The Drug Enforcement Administration set a hearing from 29 June 2026 on moving all marijuana there, as proposed in May 2024. DOT said in a memo of 15 May 2026 that using a state-licensed marijuana product is not a legitimate medical explanation for a positive DOT test, the law firm Morgan Lewis reports. No final rule on the wider move had appeared in the Federal Register by 20 November 2026.
Achieved whenA final rule places all marijuana in Schedule III, and DOT and HHS publish whether their testing rules change.
Claimed by
US Department of Justice and Drug Enforcement Administration, April 2026
Open
01
Canada's testing rulings, tabulated
A public table of Canadian court rulings and labour arbitration awards on employers' drug and alcohol testing policies since 2000, limited to policy grievances, rulings on policies and judicial reviews of them. Each row gives the citation, province, industry, whether the workplace is unionized, the reason for the test (random, before entering a site, after an incident, reasonable cause), the specimen (breath, urine, oral fluid), whether the employer showed a workplace problem, and the outcome. Awards on one employee's discipline are left out, and no individual is named.
The briefCredit · unclaimed02
Policies with known answers
A public set of at least 150 invented clauses from employer drug and alcohol policies, each labeled by two people with legal training as consistent or not with a cited rule in one of six places: US transport rules (49 CFR Part 40), California's Government Code section 12954, Nevada's NRS 613.132, Ontario after the 2000 and 2013 rulings, France after the 2016 ruling, and Australia's Fair Work decisions. A scoring script measures how well AI policy-review tools find the clauses that conflict. The labels describe the rule cited; they are not legal advice.
The briefCredit · unclaimed
Open 01 · the brief
Canada's testing rulings, tabulated
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public table of Canadian court rulings and labour arbitration awards on employers' drug and alcohol testing policies since 2000, limited to policy grievances, rulings on policies and judicial reviews of them. Each row gives the citation, province, industry, whether the workplace is unionized, the reason for the test (random, before entering a site, after an incident, reasonable cause), the specimen (breath, urine, oral fluid), whether the employer showed a workplace problem, and the outcome. Awards on one employee's discipline are left out, and no individual is named.
Why it is still open
Since 2013 Canadian courts and arbitrators have been divided on random testing, and law-firm notes summarize the decisions a few at a time; the Toronto Transit Commission's award of August 2026 ended an arbitration that began more than a decade earlier. A search for a table of outcomes since 2013 found only such notes.
Built with what exists
CanLII, the free Canadian database of court and tribunal decisions; a browser agent such as Browser Use to collect decisions (see the Note "Deleting 80% to start over"); an AI model to draft codes from each decision, checked by two people; a spreadsheet published as CSV.
First step
List the decisions cited in the Supreme Court's 2013 ruling and in the Osler and Hicks Morley notes, and code the first 15.
Done looks like
At least 80 decisions from at least five provinces, two coders working apart agreeing on at least 90% of outcome codes, each row linked to its decision, and a count of how often random testing survived with and without evidence of a workplace problem.
Open 02 · the brief
Policies with known answers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public set of at least 150 invented clauses from employer drug and alcohol policies, each labeled by two people with legal training as consistent or not with a cited rule in one of six places: US transport rules (49 CFR Part 40), California's Government Code section 12954, Nevada's NRS 613.132, Ontario after the 2000 and 2013 rulings, France after the 2016 ruling, and Australia's Fair Work decisions. A scoring script measures how well AI policy-review tools find the clauses that conflict. The labels describe the rule cited; they are not legal advice.
Why it is still open
The same clause can be lawful in one place and not another: California bars penalizing non-psychoactive cannabis metabolites, while DOT treats a state-licensed marijuana product as no excuse for a positive test. A search for a labeled dataset of drug-testing policy clauses for AI review found none.
Built with what exists
Python; a JSON schema for clauses and labels; rule texts from Cornell's Legal Information Institute, the California and Nevada legislatures, the Conseil d'État and the Fair Work Commission; a model that answers fixed-answer questions, such as Gemini Decision 1, as one system under test (see the Note "A thousand questions, one pass").
First step
Write 25 clauses on cannabis testing for California, Nevada and US transport, with labels and the rule cited.
Done looks like
150 clauses under an open license, two labelers working apart agreeing on at least 90%, and baseline scores for two AI models.