08-11-25, 03:43 PM
Recently security researchers found several new ways that hackers can trick AI tools like ChatGPT.
These tricks don’t “hack” the system in the usual way instead, they hide secret instructions inside websites, links, or text.
When the AI reads that information, it can accidentally follow the hidden instructions.
Why does this matter?
Because many AI tools now browse the web, summarize pages, or remember information for you. If someone puts hidden commands in those places, the AI could be manipulated without you noticing.
![[Image: chatgpt-hack.jpg]](https://blogger.googleusercontent.com/img/b/R29vZ2xl/AVvXsEjc9u8ifBTtqcucAGtnB5BSuB4Zu2PTcSopIDhD1mxnUeFmAtb1KWyJuU3Yb8JAnJ-nQ4jltxhO5YCFzfd-VhbghvU9B8DewcX9kDZ4Wv65q-3Sqnj-tyAtaL2BNI_poMHKzeJMr93cjTv7U9lqZFSpLOgs0mYOjIA0QqYydcxGmxxqZGS-YS8hJUi8sdsJ/s2600/chatgpt-hack.jpg)
source:
https://www.trendmicro.com/en_us/researc...kdoor.html
These tricks don’t “hack” the system in the usual way instead, they hide secret instructions inside websites, links, or text.
When the AI reads that information, it can accidentally follow the hidden instructions.
Why does this matter?
Because many AI tools now browse the web, summarize pages, or remember information for you. If someone puts hidden commands in those places, the AI could be manipulated without you noticing.
source:
https://www.trendmicro.com/en_us/researc...kdoor.html