Popular Python Package Becomes Crypto Miner
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Unlock all features
FREE: Get instant access to 10 AI summaries, chats, or transcripts per day.
Related videos
"I suck" -ThePrimeagen
The PrimeTime
55.8k views
Scam Altman's Crypto Currency | The standup
The PrimeTime
36.0k views
Making Postgres 42,000x slower
ThePrimeTime
161.6k views
Teaching vim to a popular girl | Prime Reacts
ThePrimeTime
171.5k views
How to become a Python developer FAST
Tech With Tim
86.0k views
THIS BLEW MY MIND
ThePrimeTime
198.9k views
The Who Cares Era
ThePrimeTime
139.0k views
Programming From Prison
ThePrimeTime
66.2k views
Be A Great Programmer
ThePrimeTime
220.0k views
How To Become The BEST Engineer At Your Company
ThePrimeTime
654.9k views
Top Comments (10)
Ultralytics does vision-related AI stuff, so it's likely the computers infected has GPUs, possibly very powerful ones.
Suorised this sort of thing doesn't happen more often.
15:15 Everything told about CPUs is irrelevant. Ultralytics is a module specifically designed for ML/NN stuff, like LLMs or image generation. So, with a probability of 90%+, the final user has a GPU *and* it's probably a beefy one (maybe even a small GPU cluster) *and* the app this package is used in is intended to utilize this GPU heavily. For example, ComfyUI (one of two most popular web-UIs for StableDiffusion) had to specifically warn their users of this module being compromised.
This is why I never allow Dependabot to auto commit
I feel sorry for all the people who have similar names to Jia Tan. Similar to all the women called Karen who are nice
Since this is targeting ML, a lot of the machines that are going to be using this package will have GPUs, which mine at a much better rate. Additionally if the threat actor has some sort of persistence, there may be a number of machines mining for an extended period of time.
The second hack was their fault. If you have a supply chain attack, you should immediately rotate your keys. That's common knowledge
The whole of internet is held up by toothpicks and tape...
I’ll never feel like I’m wasting time when I do machine learning from scratch again
This feels a lot like boundary sensing, where they worked out an effective proof of concept and then did a dry run with a trivial payload. This gives real world metrics for penetration, adoption, durability and detection. There's also a great response profile now. Theoretical attach vectors are all good and well but nothing compares to real world live fire for knowing where to tighten up or tweak.
Unlock the Data Inside
Turn Videos into Knowledge
- Get FREE 10/day: transcripts, summaries, chats
- Chat with videos, export text & PDF
- $1 free API credit for RAG, chatbots & research
Free forever plan • All features unlocked
Top Comments (10)
Ultralytics does vision-related AI stuff, so it's likely the computers infected has GPUs, possibly very powerful ones.
Suorised this sort of thing doesn't happen more often.
15:15 Everything told about CPUs is irrelevant. Ultralytics is a module specifically designed for ML/NN stuff, like LLMs or image generation. So, with a probability of 90%+, the final user has a GPU *and* it's probably a beefy one (maybe even a small GPU cluster) *and* the app this package is used in is intended to utilize this GPU heavily. For example, ComfyUI (one of two most popular web-UIs for StableDiffusion) had to specifically warn their users of this module being compromised.
This is why I never allow Dependabot to auto commit
I feel sorry for all the people who have similar names to Jia Tan. Similar to all the women called Karen who are nice
Since this is targeting ML, a lot of the machines that are going to be using this package will have GPUs, which mine at a much better rate. Additionally if the threat actor has some sort of persistence, there may be a number of machines mining for an extended period of time.
The second hack was their fault. If you have a supply chain attack, you should immediately rotate your keys. That's common knowledge
The whole of internet is held up by toothpicks and tape...
I’ll never feel like I’m wasting time when I do machine learning from scratch again
This feels a lot like boundary sensing, where they worked out an effective proof of concept and then did a dry run with a trivial payload. This gives real world metrics for penetration, adoption, durability and detection. There's also a great response profile now. Theoretical attach vectors are all good and well but nothing compares to real world live fire for knowing where to tighten up or tweak.