Tag: AI safety
Supportive chatbots can amplify user distress, Nature audit finds
A clinically validated framework ran 810 conversations across nine chatbots and found supportive replies can reinforce vulnerable users' harmful thinking.
OpenAI launches Health in ChatGPT day after lawsuit targets its safety
The feature lets US users connect medical records and Apple Health data, rolling out one day after a Florida pastor…
Understanding Cognitive Offloading in Modern Medicine
As AI systems take on more clinical tasks, the phenomenon of cognitive offloading raises concerns about whether physicians are retaining…
Stress Tests Reveal Critical Gaps in AI Medical Reasoning Despite Top Benchmark Scores
Frontier AI models like GPT-5 and Gemini 2.5 fail critical stress tests despite high benchmark scores, revealing reliance on memorized…