#587186

So, one thing they do very well is they model persuasive language. And the drawbacks that prevents its industrial deployment, namely hallucinations, is a feature when you're spearfishing someone, right? So, it is asymmetric in the attacker's favor when it's used for fraud and spearfishing and so on. And once the spearfishing is succeeded, they can synthesize a malware that lives off the land, meaning that it doesn't carry any malicious payload; It just carries an AI that can write malicious attacks on the fly in the target computer, which is almost impossible to detect because previously, it had to be remote controlled.