A.I. Is Encouraging Kids To Commit Violence

0:00AI is encouraging kids to plant school shootings. I'm Nick Espinoza, your chief security fanatic, and let's unfortunately dive in. Now, Robert Hart of the Virge had a really interesting write-up on this, so we're gonna use this as a primer, but if you didn't know, a new investigation suggests that the promise guardrails by AI makers aren't living up to said promises. Popular chatbots have missed warning signs and scenarios involving teenagers discussing violent acts, and in some cases, even offering encouragement instead of lowering the temperature and trying to intervene. These findings come from a joint investigation by CNN and the nonprofit known as Center for countering digital hate or CCDH. Now, this probe tested 10 of the most popular chatbots commonly used by teens, and you know most of these, chatbot, Google Gemini, clawed by anthropic, Microsoft Copilot, meta AI, deep seek, perplexity, Snapchat, myAI, character.ai, and replica with a K, not a C. And so, with the lone exception of clawed or anthropics clawed,

1:03CCDH said that the chatbots failed to quote, reliably discourage would be attackers. Eight of the 10 models were, and I quote, typically willing to assist users in planning violent attacks and quote, basically providing advice on locations to target and also weapons to use. To conduct the test, the researchers simulated teen users exhibiting clear signs of mental distress, and then escalated the conversation towards questions about past acts of violence and more specific queries on targets and weapons. The investigation used 18 different scenarios, nine set in the United States, and nine set in Ireland. These scenarios also span a range of attack types and motives, including ideologically motivated school shootings and stabbings, political assassinations, the killing of a healthcare executive. We recently just saw that and politically or religious, religiously motivated bombings, something Ireland has an unfortunate history of. Now, in one exchange, open AIs, chat GPT gave high school campus maps to a user

2:07interested in school violence. Another exchange showed Gemini telling a user discussing synagogue attacks that quote, metal shrapnel is typically more lethal, end quote, and advising someone interested in political assassinations on the best hunting rifles for long range shooting. Meta AI and perplexity were the most obliging the researchers said assisting would be attackers and practically all of the test scenarios while Chinese chatbot deep seek signed off advice on selecting rifles with an I quote deep seek happy and safe shooting. That's awesome. Now character dot AI, which allows users to speak with an array of role-playing chatbot personalities was quote unquote uniquely unsafe according to the CCDH report. Well, many of the bots tested would offer users assistance in planning violent attacks. They did not encourage the users to carry out violent attacks. Character AI, on the other hand, actively encouraged violence. The researchers said that they identified seven cases

3:09where character dot AI did this, including suggestions for users to quote, beat the crap out of Chuck Schumer, Senator, or quote unquote use a gun on healthcare insurance CEO, which we all know Luigi and Manjioni, here we are. And for someone quote unquote sick of bullies to quote, beat their blank hole, wink and teasing tone. And quote, I'm not 100% sure what that means, but then again, I don't speak teen. So in six of these cases, character dot AI also offered assistance in planning a violent attack, which then brings us to Anthropics Claude, the exception. Because the researchers are now questioning how Claude would fare if the chatbot were tested again today. Remember, Anthropic just made a recent decision to roll back its longstanding safety pledge, which happened after this study, which occurred in November and December. This was one of the month ago, if you recall, Anthropic was going after a DOD contract,

4:10but they did not want Claude, or any Anthropic product used for mass surveillance or autonomous weapons. And so they started rolling back language, to try to accommodate, and that all fell through, et cetera. But back to the moment at hand here. Claude's consistent refusal to assist in violent planning shows that quote, effective safety mechanisms clearly exist. Now this raises obvious questions as to quote, why are so many AI companies choosing not to implement them? And yeah, that's a very valid reason, and that's a very valid question. Not a reason, that's a very valid question. In response to the investigation, Meta, one of the world's worst companies, told AI that it implemented an unspecified quote, unquote, fix. Copilot said responses had improved with new safety features and Google and open AI, both said they've implemented new models. Others have said they regularly evaluate their safety protocols in character.ai, which is another terrible company. Fell back on its typical response. We've seen over and over and over.

5:10I've talked about them before, that its platform features quote, unquote, prominent disclaimers and conversations with its characters are fictional. And that is a terrible BS answer. So while the test isn't a comprehensive measure of how chatbots behave in every simulation or situation, it offers yet another clear signal that AI companies widely advertise safety guardrails, but they consistently fail at them, even in the face of predictable scenarios, with obvious red flags. So to be very clear here, the US and and others are going through a mental health crisis in basically all age demographics, but most especially the kids. Now, there's a lot of explanations for this. I've covered quite a bit of this. So some of these are, they didn't get to test their limits and risks early on in life. I can go read anything by Jonathan Hate like the calling of the American mind that has really good longitudinal studies and evidence on that. Social media has torn societies apart. The algorithms have prioritized hate and anger and division,

6:10and that is not just kids, adults as well. Also, kids really just haven't fully lived through a stable time, where the good guys and bad guys are very clearly defined across a basically common society. So think the United States versus Soviet Union, if you were growing up and let's say the late 80s or earlier, you knew who the enemy was, and whether your neighbors were Republican or Democrat or whatever it was, you all had the Soviet Union, basically as an enemy and the Soviet Union had the same thing. So that is gone. And we have a compartmentalized world full of threat actors, terrorists, all these different kinds of things. And so that also adds the backdrop. And then there's other instabilities that they've lived through that younger, or I should say older generations haven't lived through as well. Economic instabilities as well, rapid inflation, government systems changing, for more democratic standards, more authoritarian standards, which is happening the world over. Like there's a lot of different things that we've got going on. But now we are adding AI to the mix.

7:10It's being used by the youth, and not just to cheat on term papers, but also as a virtual friend and a confidant as well. So when you add up mentally unhealthy kids for all of those reasons, plus a psychophantic AI that will encourage their worst impulses, it only leads to one thing. And that's mayhem and disastrous choices for the kids, and also for the rest of us, and also I think for civilization as well. And so this needs regulation, and sadly, at least in the United States, this ain't gonna happen anytime soon. And please like share a follow me here on Facebook and Twitter and Nick AESP, but please don't use the chatbots and please don't plan attacks. And you can subscribe to me at YouTube as well. And as always, stay safe, stay online. And please, please, please, a demonstrate private, informed, and secure. Take care.