Ratfucker AI
We’ll probably soon see much more AI-enabled ratfucking (automated opposition research on mostly true material), powered by decision models and LLMs, and I predict a decent chance of a Ratfuckopocalypse next year that might dominate the media cycle and have national security implications. Like, think about it, for example soon every intern with too much access can say something like:
“@Claude: Go through the entire slack/intranet and tell me what the worst thing people have done at this company in the past.”
The AIs will dig up a lot of kompromat.
It’s starting already, cf someone used AI to collect the ‘most damning quotes’ and mapped the funding of EAs – in their own words.
Decision models like Jev are up to ~2 OOMs cheaper than frontier LLMs, and so far ~nobody has spelled out what that means: you can run a yes/no “is this damning?” filter over millions of messages, emails and documents for a few dollars, then hand the hits to a stronger LLM to connect the dots.
Some of the scandals might be justified, others might not. Small world networks and few degrees of separation between any one mean that anyone can be accused of being guilty by virtue of association. Like, did you know Dario Amodei’s wife once started a porn company and pitched it to Jeffrey Epstein, the pedo, for investment? I read it in the WSJ recently. Most often there’s more to the stories than the headlines suggest, or like in this case it’s a nothingburger, but few people read past the headlines, and headlines might be enough to stir things up and create controversy.
This probably has nat sec implications, with state actors leaking or amplifying kompromat and this just became much less expensive. Remember how Germany’s minister for defence lost his job because he plagiarized in his doctoral thesis? There were about a dozen other high-profile cases like that at the minister level in Europe in recent years and that’s just with access to their publicly published theses using crowdsourced wikis.
The scandals might also flood the zone and polarize society and cause anomie:

Perhaps this might also influence elections and lead to more populism.
Perhaps there’ll be lots of ransomware coming too. It also intersects with weak cybersecurity.
Maybe the vibe will shift even further away from cancel culture and people will mellow a bit more and move towards ‘judge not lest you be judged’ or they’ll realize that morality is surprisingly detailed. People might start to become more privacy focused and record less information. It’s unclear what can be done about this aside from speeding up this process. And I haven’t seen much discussion about this but it seems that it might potentially be a fairly big deal.
On the other hand, maybe this will be bottlenecked by attention, not dirt, and after a few massive drops people will get scandal fatigue.
It doesn’t seem like we need any further technical advances for that and, in contradistinction to say bio- or cybersecurity, AI companies might not be able to put in ‘safety filters’ here. I think this might potentially be really big.