X @The Economist
The Economist·2025-09-26 23:00
Large language models do not separate data from instructions. If they are given a command, they will attempt to follow it. This oversight can become a security vulnerability https://t.co/m9LSYhXo5q ...