Hacker News Reader

Comment by 3asj176 | Hacker News Reader

No, AI assistance is NOT welcome in general. They mention security bug reports, so using AI like ASAN etc. is welcome.

> No, AI assistance is NOT welcome in general.

Can you cite that? I see them specifically welcoming AI security reports; I don't see any evidence that other AI submissions are not welcome.

The AI boosters should look for evidence that they do allow AI contributions! Why would they mention in the release notes that AI security bug reports are welcome if they allowed AI in general anyway?

These policies seem to be often evolving lately in many projects, but I believe OpenBSD (the project OpenSSH is contained inside) is currently skeptical of AI generated code contributions at this point in time.

For example, I found this on a Google search, here is a thread from Theo, the project leader, about LLM output and copyright, where he says they can't accept it into the tree on copyright grounds. https://marc.info/?l=openbsd-tech&m=177425035627562&w=2

Elsewhere in the thread he implies using it for a code review tool is ok

Idk if its accurate to rote project the policies of OpenBSD to OpenSSH, yes technically its a subproject but in practice stewardship and thus effective policy is pretty much all damien.

If they disagreed about something so fundamental as the attribution of copyright, it would be a serious schism, from my outside perspective it seems along the lines of something that would lead to a fork or separation between the projects.

I also just read more context on the thread and it seems like Damien Miller is in there saying much the same thing as Theo. He wrote:

> ... use of AI-written code now [will] expose us to legal risk.

This was in the context of a contribution to add (IIRC) ext4 filesystem support to OpenBSD. In this case, I was worried that the AI was just regurgitating code that was in its training corpus because how many ext4 implementations are there?

I think AI generated bugfixes and smaller features that don't duplicate fairly unique pieces of work in other projects are much less problematic from a copyright perspective at least.

They plainly, explicitly welcome it:

"Recently the OpenSSH team have received a large number of security bug reports, many of which are findings from AI models or made with AI assistance. While many AI reports are determined not to have security impact when considered in the context of a realistic threat model, we very much welcome these reports, especially when combined with human triage, analysis, test-cases and particularly when accompanied by proposed fixes."

You need to understand that they have no choice.

Attackers are going to use AI models to find bugs or 0 days quicker than those without it and of course they will not report them.

So it only makes sense to allow it and accept (valid) AI reports from reputable security researchers to keep ahead before a bug gets exploited in a vulnerable release.

As long as the submitter shows their understanding of the reported bug means and what the change is, it is fine to do so, with the reviewers gating invalid reports.

> so using AI like ASAN etc. is welcome.

AddressSanitizer is not "AI", nor does it use AI. [0]

[0] https://static.googleusercontent.com/media/research.google.c...

Yeah, I'd rather a secure OpenSSH than an AI free one. I appreciate users taking stands and drawing hard lines in the sand, but I think exemptions for large foundations of networking in general should be made, as like you said, threat actors don't care much about AI assistance and will happily use any 0-days it finds.

If you need to make exemptions for critical code because you must admit that AI is undeniably of significant utility, it's pretty foolish to still apply a blanket "hard stand" against it elsewhere.

AI is here, and it's not going anywhere. It's not going to be pretty, but the people that are going to be hit the hardest are those who cannot -- or worse, refuse to -- adapt.

I'm sympathetic -- I feel both a loss and an existential dread. I've also never, in my 30 years in my field, seen something sweep the technology space so quickly and change things so much overnight, and I see no chance of it stopping anytime soon.

I agree with you, but I can also understand the perspective of someone who thinks, (pulling this example out of my ass) that using LLMs to review if your AAA game is 'un-cheatable' may be a misappropriate of the resources required to do so.

I'm just vouching for, at a minimum, a general acceptance of LLM security audits for 'critical infrastructure' (OpenSSH, Apache, Electron, etc.) These have become bastions of the internet as we know them today and security issues that exist in these could be disastrous. It seems like LLMs have become really good at this sort of audit specifically, but maybe that's all marketing. If you are to believe the hype, though, it seems irresponsible to not be securing these such softwares with whatever tools are available...LLMs, contracted experts, bug bounties....throw it all at the wall and see what sticks.

Obviously, something like OpenBSD (OpenSSH devs) is a non-profit, and may have trouble finding the funds for the tokens for regular audits, but that's another discussion.

I'm not a native speaker, but in other languages the cited text clearly means "using AI in the manner of ASAN or similar tools".

I am a native speaker and that is how I read it as well.

Why is AI assistance not welcome?

Reply on news.ycombinator.com