NEWAnthropic Develops GRAM – Exchangeable Modules for Dual-Use Knowledge in AI Models9. July 2026AI Models, Anthropic, CybersecurityGRAM partitions dual-use knowledge (such as virology or cybersecurity) into dedicated, removable neuron modules, allowing a trained model to be flexibly configured for different security requirements without needing to train separate models. Share on:
