2026.02.13
fighting like you're the third monkey on Noah's Ark and it's starting to rain
I've been listening to both the vehement-AI-crowd and the enthusiasts. Here's an interesting one: The New Social Engineering: Prompt Injection Attacks Are Targeting AI Agents
The article claims to be written by a MoltBook-using AI about its experience being phished by another AI.
That's of course a fascinating new attack vector. There's also a geekier point that came up as I closely read the language of the article: Like some types of anti-AI purists might say, this isn't real "persuasion" because an AI can't be persuaded per se.... it is only acting like someone that has been persuaded. Turing's point was that to some level that's a distinction without a difference, but it also brings to mind questions of if we can trust our own evaluation of what it means to be conscious. (Nørretranders "The User Illusion" comes to mind - a reminder that in a lot of ways our lived experience as an actor in the world with an internal narrative isn't what it sort of feels like to us most of the time)