I used to be the Security Team Lead for Web Applications at one of the largest government data centers in the world but now I do mostly “source available” security mainly focusing on BSD. I’m on GitHub but I run a self-hosted Gogs (which gitea came from) git repo at Quadhelion Engineering Dev.
Well, on that server I tried to deny AI with Suricata, robots.txt, “NO AI” Licenses, Human Intelligence (HI) License links in the software, “NO AI” comments in posts everywhere on the Internet where my software was posted. Here is what I found today after having correlated all my logs of git clones or scrapes and traced them all back to IP/Company/Server.
Formerly having been loathe to even give my thinking pattern to a potential enemy I asked Perplexity AI questions specifically about BSD security, a very niche topic. Although there is a huge data pool here in general over many decades, my type of software is pretty unique, is buried as it does not come up on a GitHub search for BSD Security for two pages which is all most users will click, is very recent comparitively to the “dead pool” of old knowledge, and is fairly well recieved, yet not generally popular so GitHub Traffic Analysis is very useful.
The traceback and AI result analysis shows the following:
- GitHub cloning vs visitor activity in the Traffic tab DOES NOT MATCH any useful pattern for me the Engineer. Likelyhood of AI training rough estimate of my own repositories: 60% of clones are AI/Automata
- GitHub README.md is not licensable material and is a public document able to be trained on no matter what the software license, copyright, statements, or any technical measures used to dissuade/defeat it. a. I’m trying to see if tracking down whether any README.md no matter what the context is trainable; is a solvable engineering project considering my life constraints.
- Plagarisation of technical writing: Probable
- Theft of programming “snippets” or perhaps “single lines of code” and overall logic design pattern for that solution: Probable
- Supremely interesting choice of datasets used vs available, in summary use, but also checking for validation against other software and weighted upon reputation factors with “Coq” like proofing, GitHub “Stars”, Employer History?
- Even though I can see my own writing and formatting right out of my README.md the citation was to “Phoronix Forum” but that isn’t true. That’s like saying your post is “Tick Tock” said. I wrote that, a real flesh and blood human being took comparitvely massive amounts of time to do that. My birthname is there in the post 2 times [EDIT: post signature with my name no longer? Name not in “about” either hmm], in the repo, in the comments, all over the Internet.
[EDIT continued] Did it choose the Phoronix vector to that information because it was less attributable? It found my other repos in other ways. My Phoronix handle is the same name as GitHub username, where my handl is my name, easily inferable in any, as well as a biography link with my fullname in the about.[EDIT cont end]
You should test this out for yourself as I’m not going to take days or a week making a great presentation of a technical case. Check your own niche code, a specific code question of application, or make a mock repo with super niche stuff with lots of code in the README.md and then check it against AI every day until you see it.
P.S. I pulled up TabNine and tried to write Ruby so complicated and magically mashed, AI could offer me nothing, just as an AI obsucation/smartness test. You should try something similar to see what results you get.
You must log in or # to comment.
I agree with you that they have consumed far more of the internet than they let on. That scrapers are shoving just everything into these regardless of legality or consent. Its messed up. Once more if the world wasn’t just a concrete jungle this could probably be a great ubiquitous tool in a faster and safer manner than it is now.
Thanks for all the comments affirming my hard working planned 6 month AI honeypot endeavouring to be a threat to anything that even remotely has the possibility of becoming anti-human. It was in my capability and interest to do, so I did it. This phase may pass and we won’t have to worry, but we aren’t there yet, I believe.
I did some more digging in Perplexity on niche security but this is tangential and speculative un-like my previous evidenced analysis, but I do think I’m on to something and maybe others can help me crack it.
I wrote this nice article https://www.quadhelion.engineering/articles/freebsd-synfin.html about FreeBSD syscontrols tunables, dropping SYN FIN and it’s performance impact on webhosting and security, so I searched for that. There are many conf files out there containing this directive and performance in aggregate but I couldn’t find any specific data on a controlled test of just that tunable, so I tested it months ago.
Searched for it Perplexity:
- It gave me a contradictorily worded and badly explained answer with the correct conclusion as from two different people
- None of the sources it claimed said anything* about it’s performance trade-off
- The answers change daily
- One answer one day gave an identical fork of a gist with the authors name in comments in the second line. I went on GitHub and notified the original author. https://gist.github.com/clemensg/8828061?permalink_comment_id=5090233#gistcomment-5090233 Then I went to go back and take a screenshot I would say, maybe 5-10 minutes later and I could not recreate that gist as a source anymore. I figured it would be consistent so I didn’t need to take a screenshot right then!
The forked gist was: https://gist.github.com/gspu/ac748b77fa3c001ef3791478815f7b6a
[Contradiction over time] The impact was none, negligible, trivial, improve
[Errors] Corrected after yesterday, and in following with my comments on the web that it actually improves performance as in my months old article
- It is not minimal -> trivial, it’s a huge decision that has definite and measurable impact on todays web stacks. This is an obvious duh moment once you realize you are changing the TCP stacks and that is hardly ever negligible, certainly never none.
drop_synfinis mainly mitigating fingerprinting, not DOS/DDoS, that’s a SYN flood it’s meaning, but I also tested this in my article!
Anyone feel like an experiment here in this thread and ask ChatGPT the same question for me/us?
So… if you don’t want the world to see your work, why are you hosting it publicly?
If I copy McDonald’s site one by one for my own restaurant and just change the name, you can expect to be sued.
And yet, their site is available publicly?
Hey Elias, found some confounding info: looks like Perplexity AI doesn’t respect the methods of blocking scrapers through robots.txt so this might just be an issue with them specifically being assholes.
Couldn’t figure out how to tag you in a comment on the other post, so I’ll edit this comment in a moment with the link.
Anything you put publicly on the internet in a well known format is likely to end up in a training set. It hasn’t been decided legally yet, but it’s very likely that training a model will fall under fair use. Commercial solutions go a step further and prevent exact 1:1 reproductions, which would likely settle any ambiguity. You can throw anti-AI licenses on it, but until it’s determined to be a violation of copyright, it is literally meaningless.
Also if you just hope to spam tab with any of the AI code generators and get good results, you’re not. That’s not how those work. Saying something like this just shows the world that you have no idea how to use the tool, not the quality of the tool itself. AI is a useful tool, it’s not a magic bullet.
And this is why AI needs to be banned from use. People own the things they post / place them under various licenses, and AI coming along and taking what you did is a blatant violation of copyright, ownership, trust, and is just general theft.
I am absolutely angry with the concept of AI and have campaigned against its use and written at length, many times, to every company that believes it’s allowed to scour the internet for training data for its highly flawed, often incorrect, sometimes dangerous AI garbage. To hell with that and to hell with anyone who supports AI.
It all started with this today:
Perplexity AI Is Lying about Their User Agent https://rknight.me/blog/perplexity-ai-is-lying-about-its-user-agent/
Discussion Primer: From my perspective and potential millions of others, the readme is part of the software, it is delivered with the software whether zip, tar, git. Itself, Markdown is a specifiction and can be consider the document as software.
In fact README is so integral to the software you cannot run the software without it.
Conclusion: I think we all think of readme, especially ones with examples of your code in your readme, as code. I have evidence AI trains on your README even if you tell it specifally not to use readme, block readme, block markdowns, it still goes after it. Kinda scary?
I want everyone else to have the evidence I have, Science.
I also just realized why I’m getting heat here, lawsuits.
I just gave legal cause that practice was not properly disclosed by Microsoft, abused by OpenAI, a legal grounds as a README.markdown containg code as being software, not speech, integral to licensed software, which is covered by said license.
If an entity does find out like me your technical writing or code is in AI from a README, they are perhaps liable?
The comments so far aren’t real people posting how they really feel. An agenda or automata. Does that tell you I’m over the target or what?
Look my post is doing really well on the cyberescurity exchanges. So to all real developers and program managers out there:
Recommend the removal of any “primary logic” functional code examples out of your
README.md, that’s it.PSA, Here to help, Elias
