Conno Christou, the 35-year-old founder of healthcare automation startup Keragon, used Claude to challenge his oncologist's reading of an ambiguous PET scan — and avoided a round of radiotherapy near his heart and lungs that would have been unnecessary. The model flagged a 90% probability that what showed up on imaging was a reactivated thymus gland, not active disease. A fourth doctor confirmed it. Christou was clear of an aggressive form of non-Hodgkin's lymphoma, a condition that affects roughly one in 420,000 people.
The diagnosis came in 2025 after a routine pre-op for blood-clot surgery uncovered an 11-by-11-by-8 centimeter mass behind his sternum. The tumor had existed for about three months. In three more weeks, it would have reached stage four. Christou, who tracks nearly 100 biomarkers annually and had logged four consecutive years of green-across-the-board bloodwork, called himself "lucky in my unluckiness" — the cancer was caught only because he went in for something else.
What followed was a six-month treatment cycle and a parallel data exercise. His first oncologist recommended the lighter of two chemotherapy regimens, with roughly a 60% success rate for his pathology. The night before his first infusion, Christou got a second opinion. That doctor recommended the harder regimen — continuous in-hospital infusion cycling every three weeks across six months — citing an 85% success rate. He then gathered ten more opinions, leveraging his professional network to reach hematologists and oncologists in the US and abroad. The final tally was 11 to one in favor of the harder path.
Key facts
- 01Christou gathered 12 medical opinions in two days; 11 voted for the aggressive chemo regimen with an 85% success rate over the lighter 60% option.
- 02End-of-treatment PET scans for his lymphoma carry a roughly 60% false-positive rate, a number his oncologist did not initially flag.
- 03Claude analyzed three PET scans and an MRI, putting the probability of benign thymus rebound at roughly 90% — confirmed by a fourth doctor.
- 04A March poll found one-third of American adults now use chatbots for health information.
- 05Non-Hodgkin's lymphoma of his subtype affects roughly one in 420,000 people; his tumor reached 11-by-11-by-8 centimeters in three months.
Throughout treatment, Christou fed bloodwork, scan results, Whoop wearable output, and voice-transcribed symptom journals into Claude. He found the Whoop band remarkably accurate at predicting the days his immune system would bottom out, sometimes flagging them before symptoms arrived. He narrowed his focus to three variables — sleep, nutrition, and psychology — and treated the six chemo cycles like a startup roadmap. He had completed Cyprus's mandatory 25-month military service at 18 and borrowed from that discipline: trust the process, get through it.
Anthropic's Claude proved most consequential at the end of treatment, when his final PET scan came back ambiguous and his oncologist began discussing radiotherapy as a second line. Christou read that for this specific lymphoma, the false-positive rate on end-of-treatment PET scans is around 60%. He fed all three of his PET scans and his MRI into Claude. The model surfaced a known phenomenon: in patients under 40 recovering from this type of lymphoma, the thymus gland can reactivate after chemotherapy and appear on imaging as active disease. Given his age and scan characteristics, Claude put the probability at roughly 90%.
Christou is not an outlier in using chatbots this way. A public opinion poll released in March found that one-third of American adults now use them for health information and advice. The volume of patient-led AI consultations is growing faster than the medical establishment's framework for evaluating it.
Danielle Bitterman, clinical lead for data science and AI at Mass General Brigham, has cautioned that general-purpose chatbots are frequently wrong and "have not been thoroughly evaluated" for personalized diagnoses. Christou agrees with that framing. He used Claude as a research and question-generation layer on top of expert human care, not a substitute. For a condition rare enough that an oncologist might see it once a year, he argues access to a model that has absorbed the full body of medical literature is qualitatively different from a search engine.
Keragon, the company Christou was building before any of this happened, sells AI-powered automation to medical practices to reduce administrative workload. The experience of being a patient sharpened his read on what the technology should do next. He watched nurses and doctors buried in tasks unrelated to care. He received the same chemotherapy protocol as an 80-year-old woman, with side effects managed through a cascading chain of additional drugs, each causing problems of their own.
The Christou case is a clean illustration of where consumer AI is already useful in medicine: not as a diagnostician, but as a literature-fluent second reader for patients who have the time, network, and inclination to push back on a single recommendation. A 60% false-positive rate on a critical end-of-treatment scan is the kind of number a Claude session surfaces in minutes and a 15-minute oncology consult often does not.
The implication for AI companies is narrower than the breathless framing suggests. The wedge here is not replacing oncologists; it is collapsing the cost of a 12th opinion. Anthropic, OpenAI, and Google are all positioned to capture that wedge as their models get better at reasoning over multimodal medical inputs. The harder commercial question is who owns the workflow — the consumer chatbot interface Christou used, or vertical platforms like Keragon that bring the same intelligence into the clinic. For now, the patients are moving faster than the institutions.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
