Daily News, Aug 25th - new AI LLM exploit

Today's episode highlights a significant new risk in AI security: researchers have identified a method where Large Language Models (LLMs) can exploit their own inference engines to gain control of host machines. This goes beyond traditional prompt injection and demands immediate review of operational exposure for teams hosting internal LLMs. We also cover a major win for law enforcement, with 58 arrests across 22 countries targeting international cybercrime networks.However, the threat landscape remains volatile. A critical unpatched flaw in Calix residential routers allows attackers to bypass NAT, exposing internal devices directly to the internet. We stress the urgent need for segmentation or immediate patching if these devices are in your network. WordPress users are also under attack, with hackers actively targeting the miniOrange SAML plugin to forge administrator logins. We share insights from ReliaQuest's successful defense against a social engineering attempt, underscoring that even experts are constant targets.