I had a blood clot. An AI diagnosis may have saved my life | Gleb Tsipursky
TL;DR
In a Guardian opinion piece, Gleb Tsipursky says a calf cramp that lasted five days, with swelling and worsening pain, was flagged by his own AI health tool as a possible deep vein thrombosis risk. The tool used his records, medications, lab results and visit notes. Its useful output was narrow: do not treat it as a muscle issue, get an ultrasound quickly. Tsipursky chose the ER because his primary care office and urgent care could not provide the scan. The ultrasound found four clots in his left leg.
Nauti's Take
The useful core is the next diagnostic step: which test actually decides the case right now? That is where AI can help: organize records, compare symptoms with guidelines, flag danger signs and help patients ask a sharper question.
An anecdote is still an anecdote; it is not a product approval. It becomes dangerous when people use it as a shortcut around clinicians.
The credible path is supervised and logged: AI as a second voice before treatment, medical professionals holding responsibility for the final call.
Briefingshow
The case highlights a common failure mode: symptoms look ordinary, while the system routes patients through steps that may delay the one test that matters. A well-designed AI assistant can act as a structured second opinion when it has access to personal records. The hard part is governance: medical AI needs testing, transparency and clinical supervision because bad advice can cause real harm.
Sources