A developer put Opus 4.8 to an unconventional test: analyzing their shoulder MRI for a second opinion. After experiencing right shoulder pain, they received a diagnosis of a Grade III partial-thickness tear requiring immediate treatment. Feeding the 266 MB DICOM files to the AI through Claude Code revealed a starkly different interpretation—an intact tendon with only mild tendinosis.

The experiment highlighted practical differences between AI interfaces. Using Claude Code instead of standard chat allowed the model to install necessary packages and run substantial analysis on medical imaging data. When the AI flagged questionable treatments—including shockwave therapy applied without calcification and a homeopathic injection—it raised concerns about the aggressive intervention plan. The author then ran an arbitration process, giving the AI both reports plus discussion notes.

The arbiter concluded with moderate-to-high confidence that no tear existed, favoring the AI's original assessment. This discrepancy demonstrates how medical AI tools can challenge clinical decisions, though trust remains difficult to establish. The experience left the author questioning whether to seek additional medical opinions or continue rehabilitation independently.

The case illustrates AI's potential for medical second opinions while exposing current reliability gaps. Technical capability exists, but validation and trust frameworks lag behind. Until then, patients face uncertainty when AI contradicts human expertise.