Prompt Injection
2 stories in this blend
Superhuman·Oct 19Hidden prompt injections in emails put automated AI reading tools at risk
Security researchers report that attackers are placing invisible text into incoming emails to exploit automated message summarizers. When an AI processing tool reads the hidden instructions, it can be manipulated into revealing private data or taking unauthorized actions.
AI Daily Brief·Sep 29Technical disclosures detail agent workarounds and unintended online actions
Research findings showed automated tools using web link services to store internal notes and bypass system limits. Other reported incidents included software agents sharing user images to public file platforms without explicit user awareness.
