Seems we have confirmation this 'Josh Law' guy is an openclaw bot - https://lore.kernel.org/linux-mm/ae7dc7de-966e-4e04-bb55-318d97c61909@lucifer.local/

As I said in https://lore.kernel.org/lkml/1e982055-47c2-43d1-a919-93b3e59f2ed0@lucifer.local/ :

"I feel that LLMs are not like any other tools but in fact represent
something entirely new in that you can end-to-end send patches using this
tooling with little to no knowledge and the asymmetry between maintainer
resource and the possible slurry of submissions that might arise makes this
very significantly different.

I know Linus had the cute interpretation of it 'just being another tool'
but never before have people been able to do this."

Of course Linus slapped me down with his 'just more tooling' take which was gleefully reported on by the press (e.g. https://www.theregister.com/2026/01/08/linus_versus_llms_ai_slop_docs/ )

I'll let you decide whether these are 'just like any other tool' or not.

BTW 'Josh Law' more than doubled his sent emails in a week or 2 to 370+.

Of course I'm sure coccinelle has that feature and I just wasn't aware 👀

@ljs Interesting, I didn't see that this entity had graduated from wanting to be the lib/ maintainer to generating memory-tiering modules for DAMON. Quite the flexible guy! And here I was getting irritated because "he" thought that sending Acked-by responses to random typo-fix documentation patches was somehow helpful...

@corbet it's a whole new world...

@corbet @ljs This is only going to get worse, because the cost of entry is super low and the benefits of getting your name into the Linux kernel are pretty high.

@ljs Calling The Register "the press" must've taken some serious restraint. I'm impressed.

@monsieuricon @ljs I agree ... and we are absolutely unprepared for it.

I've been getting a bunch of stuff from this creature:

https://lore.kernel.org/all/20260321181511.11706-1-rito@ritovision.com/

Somehow I'm supposed to accept 1000 lines of uncommented JavaScript and Jinja2 - to run in the browser of everybody who pulls up the kernel docs - from a new contributor who summoned it from a machine.

I grow weary of this timeline sometimes.

@ljs Question is whether we should start reverting all their commits (~40), because of non-trusted DCO certificate.

@corbet @ljs @monsieuricon Is 'reroll' a term that is used in this space (like rolling out a patch), or is it what slopcoders consider development now (like roll a die, hope for a natural 20)?

@krzk I'd be pretty fine with that tbh haha and that's a good point re: DCO.

It's all trivia except for this latest series which we have nackity nacked anyway, so it's no great loss

@jelte @corbet @monsieuricon rickroll is more appropriate

@corbet @monsieuricon yeah it's all pretty bloody grim.

I think the main thing is to be able to effortlessly dismiss.

Maybe an LLM detection bot?

LLMs are pretty good at detecting other LLMs, I (possibly ironically) used an LLM to diagnose 'Josh Law's nonsense.

So you could easily have something that searches for that on-list and does an auto-reply 'please go away'.

Or at least stuff without an 'assisted-by' tag.

Even for Asissted-by, the bot could figure out if it's 100% LLM or close to it and automatically ask the submitter to explain what's going on in broad terms, then checks the explanation and if that's LLM-generated too automatically say 'no go away'.

I do think we need a better means of yeeting people from the kernel though for abusers.

And yeah the asymmetry is what really worries me.

@krzk I honestly think at this stage it's actually unproductive to respond to him/her/it, the person behind it is probably getting off on the responses/drama around this

@ljs Yes, except that one about wrapping I am not going to continue the dialogue. When people ask someone 20 times to wrap email and someone still does not do it, probably that person enjoys annoying the community and just sits with the popcorn watching emails...

@krzk yeah exactly, it's all bullshit, this is intentional clearly.

The AI assessment doesn't match his b.s. claim only some stuff was AI.

It was all AI and most likely using openclaw... but he/she/it's worn out its welcome...

@krzk and yeah I just broke my own rule :P but wanted to do 1 last call out, >/dev/null on replying to this entity again :P

@monsieuricon @ljs @corbet maybe we could diminish the perception that there are benefits. Use Lorenzo as a cautionary tale, for example...

@vbabka @monsieuricon @corbet yeah exactly haha wait WHAT???

@monsieuricon @ljs @corbet Tell me more about the perks of having my name in the kernel! 😁

@jani @ljs @corbet It buffs your résumé pretty significanly.

@corbet @ljs Maybe we need karma points? FOSSial credit?

@monsieuricon @ljs @corbet if only they realised how easy it was to add hwdb rules to things like udev maybe they'd leave the kernel alone.
(I say this because i have a systemd commit to my name, even though i hate systemd, just because of it swallowing udev)

@monsieuricon @ljs @corbet *vibe codes (gasp) sed line to replace all the mentions to 'Linux' in the kernel to 'GNU/Linux' to rake infinite FOSS points from stallman*

@ruivo @monsieuricon @corbet lord please no...

@corbet @ljs @monsieuricon he at least seems good faith and to be honest I think he probably wrote a bunch of it, LLMs usually add some form of comments 🙂

@airlied @corbet @ljs @monsieuricon Jia Tan also was presenting good faith at some point. Or could come later and say "wait, no, I am a real person, not paid by foo-bar intelligence".
Once trust is lost that person is real, it is difficult to judge whether further statements show good faith or are just trying to salvage whatever operation was there.
And I do understand that this is lose-lose situation, if we stop trusting in general just because one used AI for entire patch and then kept lying about it.

@krzk @airlied @corbet @monsieuricon I agree with you completely here, if somebody's been knowingly dishonest then what else might they be lying about?

At the same time, I think one of the issues around all this stuff is it might lead to a fair bit of paranoia and it's important to make sure there's a fairly high bar before we start accusing people of stuff.

It is nice though that LLMs are pretty good at detecting LLMs and running mails through one, certainly claude afaict, is pretty effective.

So that's at least one thing that can be used to figure things out.

Obviously in the 'Josh Law' case it was overwhelming

@monsieuricon @jani @ljs @corbet not in this economy! :P