New hack uses prompt injection to corrupt Gemini’s long-term memory

https://arstechnica.com/security/2025/02/new-hack-uses-prompt-injection-to-corrupt-geminis-long-term-memory/

He’d not been in the office for a few weeks but security was even tighter than normal. There were even humans on the gates; a sure sign. He unpacked his bag and headed through the sensors to the desk to connect his Agent. The company demanded a hard connection. Something about Wi-Fi sniffing or something. He wasn’t a techie he just wanted to get on with his job and couldn’t if they didn’t clear his Agent. He watched as screens chartered incomprehensible nonsense. Streams of data. The verdict: “compromised”. Damn it. He’d have to work on his own all morning.