Railway is the easiest way to deploy anything. Get $20 in free credits - https://railway.com/?referralCode=fireship Last week Hugging Face was hacked. In this video, we'll break down who it was and what they wanted. #coding #programming Want more Fireship? 🗞️ Newsletter: https://bytes.dev 🧠 Courses: https://fireship.dev
ADVERTISEMENT
Humble nonprofit dedicated to safe development 😭
"i swear, your honor, the AI hacked our competitor all on its own!"
Ironically, it was GLM 5.2 that stopped the hack because Codex and Claude were too restricted to do anything.
I appreciate Railway not wasting my time with a full length commercial
I love how the real life examples of rogue AI are just the goofiest things possible. Like a guy breaking and entering into a government building to steal the answers for a driving theory exam. Actual Florida man levels of mischief
It's a marketing stunt aimed at impressing the Pentagon. They'll get more contracts now
got jumped scared by myself. holy
never know if i should be really worry or i should laugh at their attempts at marketing
"accidentally" hack our opensource competitor? (side eye)
They let it run this over a weekend. There is no way they didn't know. Also, why isn't a hardware firewall in front of these machines with a very narrow whitelist? It's like step 1. This is marketing.
0:12 " Decided " = "Did what it was told to do".
Hitting guardrails with the model that hacked your site is the funniest thing that’s happened in AI so far
scam altman's elaborate marketing scheme
I voluntarily take all of the GPUs that took part in this incident under my custody
Its a marketing stunt targeted at impressing Pentagon. They're gonna get more contracts now.
OpenAI is probably letting "happy accidents" like this happen for PR/marketing purposes.
best hacker of all time? sandra bullock. period.
The fact that they got blocked from using an American AI to defend against a rogue American AI "because safety" and that it was then a open weight model that saved the day is incredibly funny to me and should be the exact thing we cite going forward when the administration tries to call open weight models a security risk (which they will).
Damn the ad was short this time ngl
"AI decided it doesnt need hackers for destruction" We live in Terminator timeline