I'M WORRIED ABOUT A PROMPT INJECTION WORM

Summary from the TLDR InfoSec newsletter

A prompt-injection worm could exploit AI agents parsing external inputs, such as emails, to exfiltrate data and propagate through victims' communication channels, so organizations should map AI integrations, model access levels, and build AI-specific incident response plans.