This isn't news this is a common occurence that has been reported in major newspapers, journals, and is a common conversation in industry conversations.

There are bad things on the internet. A tool that effectively summarizes the internet (due to the way it was built), is going to contain bad things.

LLMs have helped people complete succesful hacks. LLMs can teach a high school drop out how to synthesize unscheduled drugs that are more addictive than the heroin and cocaine we all grew up hearing about. They can also teach synthesis of the "boring" stuff like LSD, meth, cocaine, DMT, etc.

LLMs can demonstrate the production of classical explosives, nuclear explosives, and hybrid dirty bombs, including the entire refinement process required to take radioactive rocks and concentrate them in to effective nuclear explosives.

LLMs can aid in doxxing, cyber-bullying, stalking, revenge porn, and "rule 34" type content.

All of this is possible because all of this was already possible thanks to the most effective communication device we as humans have ever created. LLMs are just a novel method of accessing the content that was already there.

All of this is capability is predated by yet earlier development reffered to as "freedom of speech" in which several countries have decided that people are allowed to say things that could be utilized in harmful activities.

There's a reason the monopolies and royalists of old warned that free speech was a slippery slope. This is that reason.

You can't have both the freedom of speech and santized speech. If you wanted sanitized speech such that such dangerous information could not be disseminated, and could not then be posted to the internet, and could not then be synthesized in to a super slick LLM search engine, you would not support freedom of speech and would then also not work at a private sector newspaper.

In fact, by your having voiced this reality to the vastly larger public that reads your paper than the relativiely small number of nerds and early adopters that already knew such information was available and served by LLMs, you have only increased the potential use of such information. Something about the Streisand effect. Because what you've neglected to share, or perhaps realize, is that there is nothing at all that the companies you've mentioned can do to mitigate this effect. There are lots of models that display the effect you've described, and the companies you've mentioned represent but a minority of them. Were anthropic et al perfectly sanitized, deepseek et al would be unaffected.