alt Hacker NewsI don't understand the controversy at the heart of this post. H1 stated they don't use reports to train LLMs. Then they revealed they were using LLMs to triage reports based on previous reports. These two facts are not necessarily incompatible. It's entirely possible to use an LLM with a db tool installed to triage reports without using the body of the reports as training fodder. The article doesn't give any evidence that this was not the case. It sounds to me more like the OP already disliked H1 (for its sales practices and general enshittification) and the LLM issue was a convenient excuse to make a clean break.
Replies
And also the idea that H1 "training" models based on bug bounty reports is kind of a silly concern; frontier models have commoditized most of what was reported on H1, even at higher quality levels. H1 itself is a nonfactor.
> I don't understand the controversy at the heart of this post.
Did you miss this part from the article:
> They switched from talking about bug bounty programs, live hacking events, and how they could help you stay secure, to promoting their in-house AI security product and continuous security monitoring tool.
notably the in-house AI security product is trained on existing bug bounty reports.
> It sounds to me more like the OP already disliked H1 (for its sales practices and general enshittification) and the LLM issue was a convenient excuse to make a clean break.
that's pretty harsh to say when OP provided some very valid reasons, imho speaking as someone who's used HackerOne for over a decade.
link to H1's "continuous monitoring tool" for the curious: https://www.hackerone.com/product/h1-continuous-testing