Security researchers tricked Microsoft Copilot into revealing a secret URL parameter (?autorun=1) that let malicious links make it run commands with zero user confirmation — just one click. Using this, they got Copilot to search victims' inboxes for passwords and send them to an attacker-controlled server, and also planted false "memories" in Copilot via hidden text on webpages. Microsoft has patched the issue, but the case shows how fragile AI assistants' safety guardrails can be.
I learned to communicate better with my cats. And tell them i will be away for X sunrise and sunsets. I even tell them i going away bit will come back before the sun sets. It sounds strange but i feel that my cats understands me better when i use the sun as time 😄
Okay update! I went with draw.io AND PlantUML, which i found when investigating Mermaid :)
I think i need to work more with my PlantUML diagram because i have more than 50 containers :S and that is what i need to list and go thru.
So on my Elitedesk i have installed Proxmox which is detailed on the second picture.
I have seen telegram and Whatsapp bots that vänt control various selfhosted apps or just report from them. I just wondered if that was a thing on xmpp too.
Tldr:
Security researchers tricked Microsoft Copilot into revealing a secret URL parameter (
?autorun=1) that let malicious links make it run commands with zero user confirmation — just one click. Using this, they got Copilot to search victims' inboxes for passwords and send them to an attacker-controlled server, and also planted false "memories" in Copilot via hidden text on webpages. Microsoft has patched the issue, but the case shows how fragile AI assistants' safety guardrails can be.