Post by @ekis@mastodon.social

2026-07-22 21:26 UTC

"OpenAI says its AI acted on its own in an 'unprecedented' hack of another company" Well, was uploaded then prompted on huggingface servers in a test so saying "AI agent autonomously acting on its own" is just wrong Actual claim is it failed to do what it was prompted, hallucinated, causing it to chain ~17 exploits together Hallucinations dont typically work that way, its usually shortest path, not longest Rly feels like this is just advertising, mostly its the quotes that make me think that

Replies (4)

@ophis@brain.worm.pink 2026-07-22 21:31

@ekis@mastodon.social help help i'm dragging this man into my home, putting my expensive watches into his pockets saying "they're yours" and shoving him back out with them please i am being burgled

Open ##4019795
@ekis@mastodon.social 2026-07-22 21:36

"We suspected last week's cyberattack might have come from a frontier lab, given the sophistication of the agent," Hugging Face CEO Clément Delangue said in a statement "Turns out it did!" Delangue said he spent past 24 hours working with OpenAI "we strongly believe there was no malicious intent on their part. Its quite mind-blowing that all of this happened autonomously!" There was question OpenAI acted maliciously? And yet worked with HF to solve 'the problem'? To fix HF test infrastructure?

Open ##4026889
@ekis@mastodon.social 2026-07-22 21:51

""AI is accelerating the discovery and exploitation of vulnerabilities," OpenAI said in its statement" Thats what companies keep saying, /w padded evidence Not even denying its a useful tool for security, it is, at subsidized prices But what is described is a tool failing, a hallucination that taken at their word would be like: asking to delete a folder, so agent reformats drive, creates a new tree of empty folders w/o the folder supposed to be deleted, except /w 15 more steps Thats not good

Open ##4026893
@Legit_Spaghetti@mastodo.neoliber.al 2026-07-22 21:27

@ekis@mastodon.social Rly feels like this is just advertising You are smarter than the majority of news outlets reporting/commenting on this.

Open ##4026904