深潮TechFlow|11月 17, 2026 00:24
[OpenAI: Astra Receives Safety Rating in 99% of Evaluated Medical Receiving Tasks] According to a report by The Defiant on November 17th, OpenAI stated that in its evaluation of AI agents operating electronic health record software, GPT-6 Astra received a 99% clinical safety rating for 325 medical receiving tasks.
Evaluators examined the model's operations and the medical records, prescriptions, and patient messages it generated, determining whether errors, omissions, or operations could potentially cause harm to patients. OpenAI's Karan Singhal stated that in cases where a safety rating was not received, the model may have sometimes acted without proper approval or authorization.
The report notes that this percentage reflects the company's assessment of task performance, not actual patient outcomes or independent clinical validation. OpenAI states that medical computer operation capabilities are available through APIs, ChatGPT for Healthcare, and ChatGPT for Clinicians.
Share To
Timeline
HotFlash
APP
X
Telegram
CopyLink