Artificial intelligence chatbots, wielded by clinicians, staff and patients in medical settings, have surged to the forefront of patient safety threats for 2026. ECRI, the independent nonprofit dedicated to healthcare technology safety, crowned misuse of these tools as the No. 1 health technology hazard in its annual ranking, released January 21. Tools such as ChatGPT, Claude, Copilot, Gemini and Grok generate responses that mimic expert authority but lack validation for clinical use, often delivering fabricated or dangerous advice.

More than 40 million people query ChatGPT daily for health information, per OpenAI data cited by PR Newswire. Yet these large language models predict word patterns from training data rather than comprehend medical context, confidently issuing errors like incorrect diagnoses, unnecessary tests or endorsements of substandard supplies. In ECRI tests, one chatbot approved placing an electrosurgical return electrode over a patient’s shoulder blade—a placement risking severe burns.

“Medicine is a fundamentally human endeavor. While chatbots are powerful tools, the algorithms cannot replace the expertise, education, and experience of medical professionals,” said Marcus Schabacker, MD, PhD, ECRI president and CEO, in the organization’s press release. This hazard eclipses even cybersecurity vulnerabilities and system outages, signaling AI’s rapid encroachment into care delivery.

Hallucinations in the Clinic

ECRI’s report details how chatbots “hallucinate”—fabricating details such as nonexistent body parts—while sounding trustworthy. MedTech Dive notes OpenAI’s revelation that over 5% of ChatGPT messages concern healthcare, with one-quarter of its 800 million weekly users posing medical questions. Clinicians tap them for quick references; patients seek self-diagnosis amid rising costs and clinic closures.

These unregulated systems amplify biases from flawed training data, distorting advice along racial, socioeconomic or gender lines. “AI models reflect the knowledge and beliefs on which they are trained, biases and all,” Dr. Schabacker warned. “If healthcare stakeholders are not careful, AI could further entrench the disparities that many have worked for decades to eliminate from health systems.” Such inequities could widen as access to professionals wanes.

The peril intensifies in high-stakes scenarios: a nurse consulting a bot on drug interactions or a patient following dosing instructions. ECRI urges verification against trusted sources, but over-reliance erodes clinical judgment.

From Oversight Gaps to Top Threat

AI risks have climbed ECRI lists steadily. Insufficient governance ranked fifth in 2024; broader AI perils topped 2025. Now, chatbots claim the pinnacle, per Healthcare IT News, due to their ubiquity and subtlety. Unlike overt device failures, erroneous advice slips undetected into decisions.

ECRI’s 18th annual list, drawn from incident probes, databases and device testing, prioritizes by severity, frequency and preventability. No. 2: unpreparedness for “digital darkness”—sudden blackouts of electronic records. Others span falsified drugs, tubing misconnections and legacy device cyber gaps. Full roster: 3) Substandard products; 4) Diabetes tech recall lapses; 5) Syringe/tubing errors; 6) Unused perioperative meds tech; 7) Faulty cleaning guides; 8) Legacy cyber risks; 9) Unsafe workflow tech; 10) Sterilization water issues.

Chatbots’ ascent reflects explosive adoption. DotMed echoes ECRI: “While chatbots are powerful tools, the algorithms cannot replace the expertise… of medical professionals.” Dr. Schabacker stressed disciplined oversight in multiple outlets.

Governance Imperatives

Mitigation demands structure. ECRI prescribes AI governance committees, clinician training and routine audits. “Realizing AI’s promise while protecting people requires disciplined oversight, detailed guidelines, and a clear-eyed understanding of AI’s limitations,” Dr. Schabacker stated. Health systems must treat chatbots as aids, not oracles.

Providers should inventory uses, set policies banning unverified clinical input and integrate safeguards like response flagging. Patients, too, bear responsibility: cross-check bot outputs. ECRI’s January 28 webcast dissected these tactics, underscoring urgency as adoption balloons.

Regulatory voids persist; chatbots evade medical device scrutiny. Yet precedents loom: AI governance topped prior lists, hinting at incoming mandates. Modern Healthcare flags this as the paramount safety risk entering 2026.

Broader Hazard Array

Beyond AI, ECRI spotlights persistent threats. “Digital darkness”—think ransomware or power failures—ranked second, exposing over-reliance on digital records. Facilities drill infrequently, risking chaos. Substandard drugs, often imported, claim third amid supply strains.

Home diabetes devices pose recall risks; slow adoption of safer connectors like ENFit/NRFit invites misconnections. Perioperative med tech underuse invites errors; vague cleaning instructions breed infections. Legacy devices lure hackers; new implementations spawn hazardous workflows; poor sterilization water contaminates instruments.

These interconnect: AI could query flawed data during outages, compounding harm. ECRI’s rigorous vetting—via labs in North America and Asia-Pacific—cements its influence; it’s a U.S. Agency for Healthcare Research and Quality Evidence-based Practice Center.

Path Forward

Hospitals like those partnering ECRI via its Patient Safety Organization (post-2020 ISMP acquisition, 2024 Just Culture buy) lead reforms. Audits reveal gaps; training builds skepticism. As costs climb and closures mount, per PR Newswire, unchecked bots fill voids perilously.

ECRI’s executive brief, downloadable free, arms leaders; full reports guide members. Industry insiders eye this as a pivot: harness AI’s efficiencies—administrative streamlining, education—while ringfencing care decisions. The stakes? Patient lives amid tech’s double-edged advance.

Dr. Schabacker’s call resonates: Balance innovation with vigilance to avert a cascade of unseen errors.