Recognizing characteristics of harm via binding affinities to human cells. Sequences flag potential risks. Leverages Confidential Computing for biology:
- System only outputs risk categories (safe/monitor/hazard) without retraining actual genetic sequences that could be weaponized