~3days' warning on a Category 5, called at Category 1
Because of AI: a registry of measured AI benefits to people
Because of AI, the Hurricane Center called a Category 5 landfall in Jamaica about three days out, while Melissa was still a Category 1.
Caveat: There is no lives-saved number and cannot be; the measured outcome is forecast skill, and the final call was made by human forecasters using the model.
~3days' warning on a Category 5, called at Category 1
Because of AI, there were an estimated 841 fewer injury-causing crashes across 270 million driverless miles.
Caveat: Company-produced, modeled comparison against an estimated human benchmark with an adjustment for roughly 32% underreporting of human injury crashes; not a randomized comparison of identical trips.
82%fewer injury crashes than human drivers, 270M miles
Because of AI, a baby girl was born to a couple who had been told there was no sperm to work with.
Caveat: One case; STAR finds sperm in just under 30% of men diagnosed with azoospermia, and larger trials are still underway.
1baby, from sperm no technician could find
Because of AI, 1,660 TB cases were found in Ethiopia between January 2025 and March 2026 in screening programs that had no radiologist to read the X-rays.
Caveat: Program report, not a controlled comparison; "contributed to" means the AI flagged people for confirmatory testing, so the counterfactual is how many would have been found by symptom screening alone.
1,660TB cases found with no radiologist
Because of AI, TB case notifications rose 80% in a tribal district of Chhattisgarh, India, including 20 confirmed cases in people with no symptoms.
Caveat: Before-and-after design in a high-burden setting; national notifications were also rising from other case-finding efforts.
+80%TB notifications, tribal Chhattisgarh
Because of AI, about a third of urgent skin-cancer referrals at some NHS hospitals got a same-day all-clear without waiting weeks for a dermatologist.
Caveat: Operational data from the vendor and trusts; at one district hospital the autonomous pathway discharged 37% of referrals and missed one basal cell carcinoma.
25,000+cancers found; 1 in 3 referrals cleared same day
Because of AI-assisted decision support, 10 fewer stroke patients per 1,000 had another stroke, heart attack, or vascular death within three months.
Caveat: Hospitals, not patients, were randomized, and the intervention bundled AI image analysis with guideline-based treatment prompts, so the AI's isolated contribution isn't separable; enrollment ran January 2021 to June 2023.
26%fewer repeat strokes, heart attacks, or vascular deaths — 10 per 1,000
Because of AI (plus tablets and pair work), junior secondary students in Sierra Leone gained about 0.26 standard deviations in math in eight weeks.
Caveat: The control arm got neither tablets nor structured pair work, so the package was tested, not the AI; results are a Google DeepMind/Fab AI technical report, not a peer-reviewed paper.
+0.26 SDmath gain in 8 weeks; about 1.2 to 1.7 years of typical progress
Because of AI, Swedish mammography screening found 29% more cancers with 44% less radiologist reading time, and 12% fewer cancers surfaced between screens.
Caveat: The trial was designed for non-inferiority; the 12% interval-cancer reduction was not statistically significant (ratio 0.88, 95% CI 0.65–1.18, p=0.41), mortality wasn't measured, and screening took place in 2021–2022.
29%more cancers found, 44% less radiologist reading
Because of AI, a man with ALS has spoken with his family and kept working from home for two years through a brain implant, at up to 99% word accuracy.
Caveat: One participant, an experimental device implanted in 2023, not commercially available; some BCIs in other ALS patients have later stopped working.
2 yrsof independent speech at home, up to 99% word accuracy
Because of AI, stroke patients in England were 57% more likely to get a thrombectomy and reached specialist centres 64 minutes faster.
Caveat: Observational, not randomized; long-term outcomes couldn't be evaluated and reasons for non-use of the AI in some cases weren't captured.
64 minfaster transfer to thrombectomy; EVT odds 1.57×
Because of AI, 30 hearing-aid users understood far more words in heavy noise — according to the company that makes the device.
Caveat: Both studies are Fortell-run with a Fortell employee as first author, the control device also uses neural-network noise reduction, and the outcomes are laboratory speech-in-noise tests rather than daily-life benefit.
31→80%words understood in heavy noise — company-run trial, n=30
Because of AI, Northumbrian Water stopped losing about 6.4 million litres of drinking water a day.
Caveat: Operational report, no controlled comparison, and the saving is attributable to the detect-and-repair package; FIDO had already delivered a 37% leakage reduction across 15 target DMAs at the same utility in 2022–23.
~6.4Mlitres of drinking water saved per day
Because of AI, physicians using ambient scribes reported modestly less burnout, exhaustion, and task load in a 238-physician randomized trial.
Caveat: Effects were modest, only one vendor saved documentation time, occasional note inaccuracies required vigilance, and the trial ran Nov 2024–Jan 2025.
238physicians, modestly less burnout (RCT)
Because of AI, 284 UCLA clinicians reported modestly lower task load and burnout from a chart summarizer — while saving no time and mostly not recommending it.
Caveat: Preprint (medRxiv, Aug 2026), self-reported outcomes, three-month window, one health system.
−22net promoter score; less task load, zero time saved
Because of AI, primary care teams at Mayo Clinic diagnosed advanced liver disease in about 5 more patients per 1,000 adults who had a routine ECG.
Caveat: The outcome is a diagnosis, not better health: whether earlier diagnosis changed treatment or survival was not measured, clinician follow-through was uneven, and the trial ran 2023–2025.
2×advanced liver disease diagnoses (5 more per 1,000)
Because of AI, 52% of students with elevated depressive symptoms who used a generative-AI coaching app for six weeks improved meaningfully, against 32% given care as usual.
Caveat: Working paper (PsyArXiv, October 2026), not peer reviewed; written by the app's maker with academic co-authors. Brief self-reported screeners (PSS-4, PHQ-2, GAD-2), unblinded, no active control, and the meaningful-improvement figures cover only elevated-symptom students with Week 6 data.
52%improved meaningfully on depression, vs 32% with care as usual
Because of AI, specialist consultations at two hospitals in western China took 3.14 minutes instead of 4.41 when patients first gave their history to a chatbot, and patients rated communication easier.
Caveat: The benefit is time and experience, not health: no clinical outcomes were measured, patients knew their arm and rated their own experience, and the control had no pre-assessment of any kind rather than a nurse-led one.
3.14 minper specialist visit after a chatbot took the history, vs 4.41 min without
Because of AI, colonoscopists using a cloud-based polyp detector found 0.82 adenomas per colonoscopy against 0.62 without it, and 43.2% of patients had an adenoma against 35.9%.
Caveat: The outcome is detection of adenomas during the procedure, not cancers prevented or later interval cancers, which the trial did not follow. The sponsor and device maker, Odin Medical, funded the trial and helped design it; endoscopists had to have an adenoma detection rate of at least 25% and all equipment was Olympus. Fieldwork was in 2023.
7.3 ptsmore patients with an adenoma found (43.2% vs 35.9%)
Because of AI, neurosurgery outpatients in Dresden who were given an AI-written plain-language summary of their own findings before the visit rated the conversation with their doctor higher, and the consultation was shorter (median 25 vs 33 minutes).
Caveat: The gain on the primary outcome is small (Hedges' g 0.33), consultation time is a secondary outcome with confidence intervals not adjusted for multiplicity, physicians rated their own autonomy support slightly lower, and no clinical outcome was measured; only the abstract could be read.
25 minmedian consultation, vs 33 min without the AI summary
Because of AI, in-hospital deaths among high-risk ward patients fell from 23.1% to 18.6% after 11 hospitals let a deterioration model page the rapid response team automatically (adjusted odds ratio 0.82).
Caveat: Not randomized and with no concurrent control: the comparison is before versus after rollout over October 2022 to August 2024, so falling post-COVID mortality or other changes could account for part of the drop; only the abstract could be read.
−4.5 pointsin-hospital deaths among flagged ward patients (18.6% vs 23.1%)
Because of AI, outpatient clinicians in a 14-hospital Boston health system who enrolled in the Heidi ambient scribe spent 1.21 minutes less on each note, though their after-hours note work did not change.
Caveat: Observational, with the authors noting possible unmeasured confounding; a minute per note is modest, burnout and patient outcomes were not measured, and the study cannot tell AI-written from clinician-written notes; only the abstract could be read.
−1.21 minnote-writing time per visit; after-hours work unchanged
Because of AI, Baptist Health in Arkansas says it has cut emergency-room boarding by about 525 patients a month, using predictive discharge analytics across its 12 hospitals.
Caveat: An executive's account in an interview, with no baseline period, no comparison group and no dates given; the change also rested on process and governance changes beyond the model.
~525fewer patients boarding in the ER each month
Because of AI-read chest X-rays, Delhi's slum screening programme confirmed TB in 877 people among 168,568 residents screened.
Caveat: A programme report with no comparison group, so it does not show how many of these cases symptom-based screening would have found; read from a news report of the paper, not the paper.
877people with TB confirmed among 168,568 slum residents screened with AI-read chest X-rays
Because of AI-assisted chest X-rays, Malaysia's lung health pilot screened 14,350 people at seven clinics and found abnormal results in 25.5%, a third of them with lung nodules now being assessed for cancer and TB.
Caveat: A ministry statement to a newspaper: it reports how many were screened and flagged, not how many cancers or TB cases were confirmed, and gives no comparison or dates.
25.5%of 14,350 people screened with AI-assisted chest X-rays had abnormal findings
Because of AI review of routine chest X-rays, Goa's public hospitals have made 18 confirmed lung cancer diagnoses and cut the average time to diagnosis by more than 50%, the state says.
Caveat: The state's and vendor's own early outcomes: no comparison group, no period given, and the baseline for the time-to-diagnosis claim is not stated.
18confirmed lung cancer diagnoses from 106,248 routine chest X-rays reviewed by AI in Goa's public hospitals
Because of an AI tool that categorizes and routes patient portal messages, the median time to first action in a Boston epilepsy clinic fell from 11.3 hours to 1.0 hour.
Caveat: A medRxiv preprint with 7 weeks of follow-up at one site and 4 weeks at the other; only the epilepsy division of two sites was studied, and time to closure did not significantly improve.
11.3 → 1.0 hmedian weekly time to first action on epilepsy portal messages, Site A
Because of AI smoke-detection cameras, Arizona Public Service says it is notified of wildfires about 45 minutes faster on average than the first 911 call.
Caveat: A utility meteorologist's statement in a news report, with no data, period or method shown; the cameras sit in remote areas where 911 calls are slow, and human analysts verify each detection.
~45minutes earlier than the first 911 call, on average, for a utility's wildfire alerts
Because of AI-watched cameras, more than half of the wildfires flagged by California's ALERTCalifornia network were detected before any 911 call.
Caveat: A secondhand trade-press summary of agency data, with no period dates and no measure of fires saved or acres averted; early detection is not shown to change the outcome, and it does not say which detections the AI made versus people.
50%+of flagged wildfire incidents detected before a 911 call
Because of AI, surgeons at two Chengdu hospitals retrieved sperm in 68.1% of men with non-obstructive azoospermia, against 47.6% with conventional microdissection, and live birth per transfer was 38.5% against 26.8%.
Caveat: Not randomized: patients were grouped by whether the AI microscope was available, the study is retrospective, and the authors say causation needs prospective randomized trials.
68.1%sperm retrieval with AI guidance, against 47.6% without
Nothing matches.