rbtfl.

security / enterprise

立場別 · 1 takes across the edition

Security industry analysis: the government's cited jailbreak involved prompting the model to read a codebase and identify software vulnerabilities, a capability Anthropic marketed as a legitimate coding feature. Argues the suspension sets a standard where any AI model that can do useful security work is a latent export-control target.

“The jailbreak cited was asking the AI to review code for vulnerabilities, the same capability Anthropic was selling as a feature.”