Conspicuous Absence

What ARE these LinkedIn bots doing?

I am not terribly active on LinkedIn but Neal Rauhauser is maintained there. I used to get periodic pretext approaches, people who knew things about me, running accounts with small numbers of connections, or otherwise just not looking right, for various reasons.

That’s shifted since the AI boom. Now I’m periodically getting these formulaic things and I wonder what they are doing.

Attention Conservation Notice:

I have no smoking gun in terms of a diagnosis here, I’m just recounting some odd AI enabled symptoms I’ve been seeing. And I’ve got some big data/big network theories on what might be happening.

Formula Derp:

I used to keep track of these when they first appeared, with an eye on them being a counter-intel problem. When I got the same approach a dozen times in a row I decided they were all junk, so I only have a couple examples.

This was a pretty Asian girls a third my age …

This one never had a profile photo.

This one was Anglo glamour shot, if I recall correctly.

Wearying Insolence:

I finally grew tired of it … both of these got clipped by LinkedIn before they could even respond.

Interpretation:

Early on in this wave of unreality I spent a long time talking to one of these things, thinking it was a very focused pretext approach. I used to get that all the time, but it’s really been … at least five years since anyone with any sense of how to approach actually did so.

So I spent a day playing rhetorical ping pong, lying subtly, offering bait, until I was sure it WAS just a really clumsy chatbot. That left me in the curious position of having to sort out why someone would spend time and money doing this.

I can think of two economic reasons …

Recruiterbots:

Someone, somewhere, has made an AI play to the effect that they can replace recruiters with AI. These were NOT the actual offering, it was a broad effort to gather enough input to train a specific system to later handle “hopeful applicants” dialog.

Those conversations are pretty specific text and those two way interactions are not something commonly available in public. I don’t have much more to say on this angle.

Network Crawlers:

LinkedIn used to have an API and there were fun tools that produced color differentiated graphs of the network from the last six or eight jobs you had. Unless you were a network analysis nerd circa 2012 you would not remember this. Part of LinkedIn’s security, both yours and theirs, has been to curtail access.

My account has 2,278 carefully curated connections. The first 300, when I recreated it in 2012, were people I knew IRL. The first thousand or so were people like that, or people that worked with them. When I opened it up, I did so by reading white papers, then sending the authors a connect request, after having commented in the open about it. I also prune aggressively - I have multiple associates who’ve gotten suspended, while I never do. That’s because when they have conflict and I notice, I go through and remove all the pro-Russia types I see.

LinkedIn’s API was never all that, so I keep a persona that does background check work and the like. This one is “seasoned”, it’s spent nearly a decade painstakingly building a presence and has evolved past the point where LinkedIn anti-fraud sees it as unusual. It’s had a couple phone numbers, it’s moved a couple times, always places where I’ve spent at least some time, it periodically interacts, and the sprint from 342 to 642 connections was all done with LIONs - LinkedIn Open Networkers.

This account focuses on fraud schemes, like OnPassive, and on LIONs based in markets where I need to have a look at something. I used to spend one afternoon a week making it look lived in, but that kinda trailed off in 2024. Recent policy changes make me want to have a “kid” who gets his first debit card, so I can fund a Pro account there, but I don’t have any specific use case, so this remains theoretical.

Network Analysis Angle:

So I suspect that these AI things are trying to run a whole of LinkedIn access game with these objectively poorly done personas. And/or maybe it’s a China long play, as they expect our economy to degrade and this is recruiting shuffle.

When Microsoft bought LinkedIn some years ago, the big thing I remember was the “superhub pogrom”. If you had more than 30,000 contacts, they’d start at #1 and cut off your network at 30,000. There are, or at least were, things like this on Twitter, but the scale was enormous.

Back when I used to run Twitter streaming data, I had an ArangoDB table named “useless”, and it was created for this guy. He had about six million followers at the time, was following two million of them back, and his presence or absence in a graph meant nothing. I’m not sure what happened here, looks like maybe post acquisition tuning put an end to his game.

There were roughly 1,300 accounts in the “useless” table. As an example, I love to see Ariana Grande do impressions of other singers, but I never want her account in a social network analysis graph. The superhubs that benefit from preferential attachment are poison when you’re trying to see things at a human scale. So in any Gephi graph creation, that table was used to winnow things that might matter culturally, but were irrelevant in terms of who talked to whom.

So these seemingly poor quality bots might still be piling up important data from the people who DO engage with them.

Worst Case:

Do you know much about the Office of Personnel Management intrusion? Chinese hackers got into this agency and made off with every SF-86, the form used to vet people for clearances. The only agency that didn’t use this was the CIA, who insisted on an internal process.

Are you thinking “Yay, the CIA is still safe!”??

Oh my sweet summer child …

Every embassy in the world has some CIA field people on staff. So go to an embassy party, meet people, get their cards … then check that stolen data. Anyone without an SF-86 is CIA. GAME OVER.

If you don’t think in terms of networks, sometimes explaining this sort of thing sounds kinda like one’s tinfoil hat is misadjusted. The OPM intrusion’s effect on the CIA was awful for national security, but it’s a gift in terms of the simple cautionary tale it provides.

This is what was on my mind when I picked Conspicuous Absence as the title. Sometimes negative information is far more useful than its positive counterpart.

Conclusion:

This post is, I guess, a companion to last week’s WWWord Salad, which was about the negative effect of AI LLMs on … pretty much everything. These AI personas I very much suspect have a network visibility angle, in addition to whatever other objectives their creator(s) might have.

Having recently been inducted into Cicada 3301, the only thing that’s public thus far is that there’s an unbranded steganography project. Given how much time I’ve spent building and analyzing networks, you would be right to guess that there is something else brewing there. I haven’t put pen to paper yet, but I did just clean up the Proxmox machine hiding in the corner, and I installed the ArangoDB repositories.

Because I see that there’s now an ArangoDB GraphRAG implementation … combine that with 300 very sharp minds on the Cicada 3301 Discord … who knows what sort of emergent phenomena might arise.

Asked ChatGPT to do a Missing Moron formation. I got language policed …