Misleading title. The entry point for the attack was a spiked Homebrew install that the dev fell for. AI tools were in the exploit chain but it’s not like slop code opened some big dumb backdoor.
What I got from the article is that this company had minimal access controls and a very reactive security approach. Those employees should not have been able to download or execute random software or “assignments”. As we saw, even if they caught the first download after the fact and replaced the laptop, the damage was already done. This is again my problem with everyone adopting the “move fast and break things” mentality. Security starts to be seen as a hindrance that just slows developers down.
To me it just reads like multiple supply chain attacks on the client tooling side. It just happened to be developer tools that integrated with ML models. Where’s the vibe coding part?
Yeah, that fits with the whole vibe coding concept where they don’t need to know what they are doing to vomit forth code.
Security isn’t that much of an issue if you just write an app for rating co-eds’ faces.
And some people can’t understand the backlash against Proton.
Proton is a lost cause. Their leadership are Trump ball garglers. I was looking at an alternative to completely replace Google and ended up choosing someone else because of their bullshit.
They can either change their leadership or lose me come renewal.
They saved some money on AI and they lost some money due to AI, I guess it balances out nicely. /s 🤥
The title is quite misleading (see this comment), the actual hack happened because of social engineering.
As I read it the AI tool helped identify which people to socially engineer.
That makes it a heck of a lot more likely to succeed.I don’t think that’s quite right. If the article is correct, the compromised version of Homebrew that the first employee installed allowed them to find out information about the team, which helped with the subsequent social engineering attack. No AI so far.
The subsequent piece of malware the second employee installed during the fake interview process was designed to allow external access, so the attackers were able to remotely make use of internal AI systems that could perform privileged actions and open up bigger holes in the company’s security.
In general AI makes sophisticated attacks much easier, especially if the devs aren’t even reading the code anymore.
It doesn’t need to be highly accurate or successful either. If it enhances a criminals ability to do sophisticated attacks 10x more often, their liklihood of success increases dramatically. Advanced cyber attacks are becoming a game of probability instead of skill.
