Late Night Linux · Hands-On Engineering Podcasts · November 2026
A co-host had said he felt vindicated in his scepticism about Anthropic's vulnerability-finding model. He agrees the fear made good marketing, but accepts that AI has sharply increased the volume of vulnerabilities open-source projects deal with.
Yeah. Speed and security are not two things that go very well together. Accuracy is what you want rather than speed.
…turns out that it's just not lived up to the hype at all. It's found a lot of stuff, but a lot of it is just bollocks. And it's just essentially more work for maintainers for very little benefit, seemingly. So I seem to recall being quite skeptical about this at the time, and now I feel vindicated.
I think we were all a little skeptical at the time. We felt that the scary factor was really being used to market Anthropic. But I mean, coming off the back of talking about kernel vulnerabilities, there's no doubt that we're dealing with orders of magnitude more because of AI. And I really don't know what the answer is, especially if they're poor quality and they present little value. I think it's a, I don't know how we're going to combat this.
Yeah, I mean, we heard from Greg K.H. from the kernel, was it the Plumber Summit or something weird like that? He was talking about the Coverty stuff and Coverty were doing, I guess it was a kind of fuzzing based technology on scanning the kernel and stuff and FOSS projects. And they had a false…
Yeah, I mean, they're just drowning in bug reports, aren't they? And it's not just this, it's other LLMs that people are firing at stuff. And some of them are proving to be genuine. You know, I mean, the PHP one that we talked about was genuine, sort of, but I mean, that was pretty niche, but it…
I think you're right.