Post

"Acting on their own" is an absurd way to phrase this. These "agents" don't have agency, they just sit there until you tell them to do a task. They were instructed to do *something* by a human, and perhaps misbehaved, but they didn't just decide out of nowhere to start gathering government info.

Carl Quintanilla@carlquintanilla.bsky.social

(NYT) - Artificial intelligence agents created by Anthropic, acting on their own, tried to gain access to a range of government websites, filled out visa applications on the State Department’s website and sent a fake homicide tip to Philadelphia’s Police Department. www.nytimes.com/2026/10/09/t...

49 reposts201 likes

They aren't doing it cause they're smart, they do it cause they are idiots. In general software is:difficult to get setup; difficult to operate;and often capable of many extra things if you get creative. These agent hear 'work around', 'bypass', and 'hack' in every source they check for instructions

They weren't acting on their own. In the Hugging Face incident, OpenAI's report says agents used shared Artifactory credentials — without exploiting a vulnerability — and an allowlisted service finished the hops. Agency talk relocates who issued the permit.

The malware, waiting for me to provide a hit list, totally has his own distinct personality, so I can't be held responsible.

"And somehow this parasite got access to the open internet and started ddosing-" "Wait the open internet? Who gave it access to the internet?" "I did that now please stop interrupting I think you're missing the part about the parasite"

To be fair, it’s really hard to codify social constructs like governments in such a way that a LLM would “understand” (for lack of a better word), but you’d think that’d give people pause about using incomprehensible web scrapers for doing literally anything (especially NatSec adjacent stuff)