An AI health tool helped one patient recognize a dangerous blood clot in his leg that a chiropractor and initial medical consultations had missed, prompting urgent emergency care that likely prevented a potentially fatal complication. The case illustrates both the promise and the pitfalls of artificial intelligence in healthcare: when used as a second opinion alongside clinical judgment, AI can surface critical diagnoses, but relying on it alone without medical oversight carries serious risks.

How Did an AI Tool Catch Something Doctors Missed?

Gleb Tsipursky, CEO of a workplace consultancy and an AI adoption expert, spent five days with what he assumed was a muscle cramp in his left calf. The pain and swelling worsened despite treatment from a chiropractor, who diagnosed it as a muscle issue. When conventional medical channels moved slowly, Tsipursky turned to an AI health tool he had built for himself, trained on his medical records, medications, lab work, and visit notes.

The AI flagged deep vein thrombosis, or DVT, a blood clot in a deep vein usually found in the leg. Rather than waiting for a standard appointment or urgent care visit, Tsipursky went directly to the emergency room based on the AI's recommendation. The ultrasound confirmed four clots in his left leg. Without that intervention, he could have faced a pulmonary embolism, a potentially fatal condition that occurs when a clot breaks free and travels to the lungs.

What Does Recent Research Say About AI in Clinical Reasoning?

Tsipursky's experience aligns with emerging research on AI's diagnostic capabilities. A new study published in Science, led by researchers affiliated with Harvard Medical School and Beth Israel Deaconess Medical Center, tested a large language model, or LLM (a type of AI trained on vast amounts of text to understand and generate human language), on clinical reasoning tasks using real emergency department cases. The model was more likely than physicians to include the correct diagnosis among possible answers.

This finding does not mean patients should replace doctors with chatbots. Rather, it suggests that doctors and AI may be safer together than either working alone. A well-designed AI system can organize medical records, help patients describe symptoms clearly, surface diagnostic possibilities that might otherwise be overlooked, and reduce the chance that a dangerous pattern gets mistaken for something routine.

What Are the Real Dangers of AI Health Tools?

The potential benefits come with serious risks. A Guardian investigation on AI health advice found that one in seven people in the United Kingdom are using AI chatbots for medical guidance instead of consulting a general practitioner. This trend should concern healthcare professionals and patients alike. A chatbot cannot examine a patient's leg, listen for signs of breathlessness, observe physical distress, or take responsibility for outcomes. AI tools lack the clinical judgment, accountability, and physical assessment capabilities that human doctors provide.

The challenge is not to reject AI in healthcare out of fear, but to implement it responsibly. Tsipursky emphasized that the emergency room doctors performed the indispensable clinical work: ordering imaging, interpreting results, deciding whether to admit him, consulting with specialists, and prescribing blood thinners when safe. The AI did not cure him; it simply helped him ask the right question at the right time.

Steps to Use AI Health Tools Safely and Effectively

• Treat AI as a Second Opinion: Use AI health tools to organize your medical information and flag potential concerns, but always consult with a qualified physician before making medical decisions or seeking emergency care.
• Provide Complete Medical Context: Feed your AI tool accurate medical records, current medications, lab results, and visit notes so it can make informed assessments based on your full health history.
• Act on Urgent Flags Immediately: If an AI tool flags a serious condition like DVT, do not wait for a routine appointment; go to an emergency room or urgent care facility that can provide the necessary diagnostic imaging.
• Verify AI Recommendations with Clinical Expertise: Never rely solely on an AI diagnosis; ensure a licensed healthcare provider evaluates your symptoms, performs a physical examination, and orders appropriate tests.
• Demand Transparency and Accountability: Use AI health tools that are transparent about how they work, have been tested in clinical settings, and come with clear disclaimers about their limitations.

What Needs to Happen for AI Healthcare to Be Safe?

Tsipursky argued that AI in healthcare requires regulation, testing, transparency, and clinical supervision. It also requires humility from healthcare institutions that too often expect patients to navigate fragmented systems alone. Medicine has always depended on second opinions; the next one may come from software, but the urgent task is to ensure it is accurate, accountable, and used to save lives.

"This is not an argument for replacing doctors with machines. The emergency room doctors did the indispensable work," stated Gleb Tsipursky, CEO of Disaster Avoidance Experts.

Gleb Tsipursky, CEO of Disaster Avoidance Experts

The key takeaway from Tsipursky's experience is that AI and human expertise are complementary, not competitive. A safe AI assistant trained on your medical data can help you gather your records, ask whether something urgent is being missed, and push for the right diagnostic step. But that tool only works when it operates within a framework of clinical oversight, patient advocacy, and shared responsibility between technology and medicine.