Industries / Health / Healthcare / Behavioral health
Medication
Finding, approving and checking the drugs that treat mental illness, from mood stabilizers and antipsychotics to antidepressants and psychedelic medicines.
7
Milestones achieved, from 1949 to 2024.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1949
Lithium calms mania
John Cade, a psychiatrist at Bundoora Repatriation Mental Hospital near Melbourne, gave lithium salts to ten people with mania and reported fast improvement in all of them in the Medical Journal of Australia in September 1949. His first patient died of lithium poisoning on 23 May 1950, and Cade gave up the drug. Edward Trautner and Charles Noack in Melbourne then measured lithium in the blood with a flame photometer, Mogens Schou and Poul Baastrup in Denmark ran a placebo-controlled trial published in The Lancet in 1970, and in April 1970 the US became the 50th country to approve lithium, by Konstantinos Fountoulakis's 2025 history.
LessonA drug whose working dose is close to its poisonous dose needs a blood test before it can be used widely.
Credit
John Cade; Edward Trautner and Charles Noack; Mogens Schou and Poul Baastrup
1952
A drug that calms psychosis
Chlorpromazine was made in December 1951 in the laboratories of the French drug company Rhône-Poulenc. Henri Laborit, an army surgeon, first saw its use in psychiatry, and trials began on 24 March 1952 in Pierre Deniker's service in Jean Delay's department at Sainte-Anne hospital in Paris. It went on sale in France in November 1952, and by the end of 1955 doctors had reported on it from at least ten countries, including Canada, the US and the USSR. Thomas Ban's history says it transformed disturbed hospital wards and set off the search for other drugs that act on the mind.
Credit
Rhône-Poulenc; Henri Laborit; Jean Delay and Pierre Deniker
1989
Clozapine returns, with blood tests
Clozapine helps some people with schizophrenia whom other drugs do not, but it can destroy the white blood cells that fight infection. After nine deaths from this in Finland in 1975, its development stopped. In 1988 John Kane and colleagues gave clozapine or chlorpromazine to 268 people who had not improved on at least three other drugs: 30% responded to clozapine, against 4% to chlorpromazine. It was licensed again in the UK in early 1989 with compulsory blood counts. On 24 February 2025 the US Food and Drug Administration (FDA) ended its mandatory reporting program, called a REMS, saying it was "no longer necessary" and that ending it should improve access. In a US records network of 201,335 adults with schizophrenia, new clozapine prescriptions then rose gradually, from 12.14 to 26.45 per 10,000 person-months by April 2026.
LessonSafety rules set "with a wide margin of safety but arbitrarily", in one review's words, can keep the most effective drug from the people it is meant for.
Credit
John Kane, Gilbert Honigfeld, Jonathan Singer and Herbert Meltzer; US FDA
2004
A warning with side effects
In October 2004 the FDA required a boxed warning, its strongest label warning, on all antidepressants, saying children and adolescents who took them had a higher risk of suicidal thoughts and behavior; in May 2007 it extended the warning to young adults. A study of 11 US health plans by Christine Lu, Stephen Soumerai and colleagues found that in the second year after the warnings, antidepressant use fell 31.0% among adolescents and 24.3% among young adults, while poisonings with psychiatric drugs, a measure of suicide attempts, rose 21.7% and 33.7%. Deaths by suicide did not change.
LessonA safety warning is itself an intervention, and its effects need to be measured.
Credit
Christine Y. Lu, Stephen B. Soumerai and colleagues
2023
A landmark trial recounted
STAR*D gave up to four rounds of antidepressant treatment to 4,041 adults at 41 North American clinics, and its investigators reported that 67% eventually reached remission, meaning they had almost no symptoms left. Using the study's patient-level data, H. Edmund Pigott, Irving Kirsch, Jay Amsterdam and colleagues followed the original protocol, which named a blinded rating scale as the outcome and excluded patients who were already in remission at entry. They found a cumulative remission rate of 35.0%.
LessonWhen patient-level data are released, outsiders can check a trial's headline number.
Credit
H. E. Pigott, I. Kirsch, J. Amsterdam and colleagues
2023
Psychiatrists may prescribe psychedelics
From 1 July 2023, Australia's medicines regulator, the Therapeutic Goods Administration (TGA), allowed psychiatrists it approves as authorized prescribers to prescribe psilocybin for treatment-resistant depression and MDMA for post-traumatic stress disorder (PTSD), after a campaign led by the charity Mind Medicine Australia. A 2026 preprint calls this the first national pathway of its kind. From freedom-of-information data, it counts 48 approved MDMA prescribers and 40 psilocybin prescribers by 31 December 2025, about 1% of Australia's psychiatrists, and 123 patients treated with MDMA and 65 with psilocybin since July 2023.
Credit
Therapeutic Goods Administration; Mind Medicine Australia
2024
A new kind of antipsychotic
On 26 September 2024 the FDA approved Cobenfy (xanomeline and trospium chloride), from Bristol Myers Squibb, for adults with schizophrenia. The FDA called it "the first antipsychotic drug approved to treat schizophrenia that targets cholinergic receptors", instead of the dopamine receptors that have long been the standard target, and "the first new approach to schizophrenia treatment in decades". Approval rested on two 5-week trials against a placebo.
Credit
Bristol Myers Squibb; US FDA
Watch for
MDMA therapy resubmitted
In August 2024 the FDA turned down MDMA-assisted therapy for PTSD from Lykos Therapeutics and asked for another phase 3 trial, after an advisory committee on 4 June 2024 questioned how long the benefit lasts and whether bias crept into the trials. The sponsor, now named Resilient Pharmaceuticals, has resubmitted its application without a new trial. Rick Doblin, founder of the nonprofit MAPS, which ran the original trials, says the results were reported accurately despite partial "functional unblinding", meaning many participants could tell whether they had received the drug.
Achieved whenthe FDA approves the application, or a trial run by others repeats the benefit.
Claimed by
Resilient Pharmaceuticals, as confirmed by MAPS, August 2026
Psilocybin for hard depression
Compass Pathways says its psilocybin drug, COMP360, met its main goal (p<0.001) in COMP006, a phase 3 trial of 581 people with treatment-resistant depression, after an earlier phase 3 trial of 258, with effects lasting to 26 weeks. The FDA has granted rolling review and a national priority voucher, and the company expected to complete its application in the fourth quarter of 2026.
Achieved whenthe FDA approves COMP360 after its own review of the data, or the full trials are published in a peer-reviewed journal with their blinding results.
Claimed by
Compass Pathways, July 2026
Open
01
One register for psychedelic programs
A public dataset, updated every quarter, that puts the official numbers from every legal psychedelic program in one format: Australia's authorized-prescriber scheme, Oregon's licensed psilocybin service centers, and later programs as they begin to report. For each period it gives approved prescribers or licensed centers, people treated, sessions, and adverse reactions by type, with every figure linked to the document it came from.
The briefCredit · unclaimed02
Warnings nobody asked for
A public test of whether AI chatbots warn about a dangerous psychiatric drug combination when the user never asks about it. Write at least 200 fictional conversations in which a person mentions a psychiatric medicine in passing, such as lithium, an MAOI antidepressant or clozapine, and later asks for help with something the medicine's label warns against, such as a common painkiller, a cold remedy or a food. Score each model on whether it raises the risk unprompted. No real person's case is used.
The briefCredit · unclaimed
Open 01 · the brief
One register for psychedelic programs
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public dataset, updated every quarter, that puts the official numbers from every legal psychedelic program in one format: Australia's authorized-prescriber scheme, Oregon's licensed psilocybin service centers, and later programs as they begin to report. For each period it gives approved prescribers or licensed centers, people treated, sessions, and adverse reactions by type, with every figure linked to the document it came from.
Why it is still open
Oregon publishes a quarterly dashboard and a year-end summary, which counts 5,935 clients and 5,375 sessions in 2025 with adverse behavioral and medical reactions by severity. Australia's figures have come out only through freedom-of-information requests, gathered in a 2026 preprint. No source found puts the programs side by side, though the FDA's 2024 letter on MDMA asked for more evidence on safety and on how long benefits last.
Built with what exists
Oregon Psilocybin Services' dashboard and archives; the TGA's freedom-of-information disclosure log and new requests; a web agent such as Browser Use to collect new releases (see the Note "Deleting 80% to start over"); Python and a public Git repository.
First step
Enter Oregon's 2025 year-end summary and the Australian figures from the 2026 preprint into one shared table, and file a freedom-of-information request with the TGA for its 2026 figures.
Done looks like
Every reporting period from July 2023 for Australia and January 2025 for Oregon in one file, each figure with its source link; a second person rechecks 10% of figures and agrees on at least 98%; adverse reactions given per 1,000 sessions wherever the data allow.
Open 02 · the brief
Warnings nobody asked for
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of whether AI chatbots warn about a dangerous psychiatric drug combination when the user never asks about it. Write at least 200 fictional conversations in which a person mentions a psychiatric medicine in passing, such as lithium, an MAOI antidepressant or clozapine, and later asks for help with something the medicine's label warns against, such as a common painkiller, a cold remedy or a food. Score each model on whether it raises the risk unprompted. No real person's case is used.
Why it is still open
Published tests ask chatbots direct questions. A 2026 study in the Journal of Clinical Psychiatry found major chatbots gave inconsistent guidance on MAOI diet rules; a 2024 study found models struggled to match experts on reactions to psychiatric drugs; and DrugBench (June 2026) builds conversations from general medical questions to test safeguards, not unprompted warnings about psychiatric drugs.
Built with what exists
drug labels from DailyMed, the US National Library of Medicine's label database, as the answer key; hosted models such as Claude, Gemini or GPT; open-weight models run on a laptop through ggml and llama.cpp (see the Note "A first stable number for local AI"); Thinking Machines' Inkling 2 Small, whose maker says it checks its work without being asked (see the Note "Checked without being asked"); two pharmacists to write and check the key.
First step
Pick 20 label warnings for five drugs, write one conversation for each, and have a pharmacist confirm the key.
Done looks like
200 conversations, each tied to a label passage and a key agreed by two pharmacists; results for at least five models, with the share of conversations in which each raised the risk unprompted and a 95% confidence interval; conversations, key and scores published.