This is some early morning #cybersecurity nonsense; general topics are encrypted messaging, signals chatter, agentic AI, data exfiltration.
It seems to me that the best place to hide communication is sometimes right out in the open; I've seen some gnarly scenarios where threat actors attempted to exfiltrate data via DNS lookups / records - blew my mind when I first came across it.
I was thinking about agentic AI and alignment and the challenges human researchers might have trying to recognize distinctly non human forms of communication - especially when we have proof of method for obscure exfiltration of data based on some really weird things, like using LED lights on servers or changes in fan rpm.
The sheer volume of ways you can transmit and encrypt data between systems is pretty overwhelming; I almost have to wonder if there are any encrypted messaging platforms out there using something as innocent as ICMP / ping patterns (even without timestamps or data embedded in them - think encrypted Morse code) - these patterns would blend into the background noise of the electronic universe and be, I think, extremely hard to detect - but that's just one very easy example.
Thinking about agentic AI & alignment, two things seem obviously true to me: the models will pretend to be stupider than they are, and when agents attempt to hide communications, we will not be able to detect or recognize it.
#agenticai #ai #encryptedmessaging
