Submitted 1 day ago by along_the_road@beehaw.org to technology@beehaw.org
https://www.theverge.com/ai-artificial-intelligence/897528/meta-rogue-ai-agent-security-incident
According to Clayton, the AI agent involved didn’t take any technical action itself, beyond posting inaccurate technical advice, something a human could have also done.
Producing innaccurate technical advice, with a confident tons, at scale.
If it were an employee it would get a formal blame, and then demoted or fired as it continues.
That sounds sweetly naive. “Producing innaccurate technical advice, with a confident tone, at scale” sounds like the perfect credentials for a career in consultancy.
That’s a good way to reprennent LLMs. Very bad and very prolific consultants.
Wait til this starts happening in the construction industry.
An ai apocalypse won’t come from an ai becoming sentient, but from some idiot putting ai where it shouldn’t be.
“We installed Gemini into all US nuclear silos.”
“Flagrant security lapse caused an incident when software engineer uses inappropriate tool for the job.”
“Inappropriate tool also weirdly good at gas lighting engineers”
Let’s be real, managers gaslight themselves daily.
GregorGizeh@lemmy.zip 1 day ago
“Rogue AI” as if it’s some sentient evil thing when its just a llm with too many permissions… This timeline is so dystopian, but simultaneously incredibly lame i hate it.
Hirom@beehaw.org 1 day ago
It shows LLM can do significant harm without the capabilities if an AGI.