Zaštitne ograde koje su postavile AI kompanije zasad funkcioniraju – ali pitanje je koliko dugo.
Ma da, to je zbog zastitnih ograda, ne zato sto modeli to ne mogu bolje. How yes no.
In April 2026, Anthropic's advanced AI model, Claude Mythos Preview, demonstrated the ability to autonomously discover and exploit a 17-year-old critical vulnerability (CVE-2026-4747) in the FreeBSD operating system, granting unauthenticated root access. The AI identified the flaw in the RPCSEC_GSS authentication handler and developed a working exploit in roughly four hours.
FreeBSD is widely considered one of the most secure and robust operating systems, trusted by companies like Netflix, WhatsApp, and Sony. While it offers advanced security features like Jails, MAC, and ZFS, it is generally ranked just behind dedicated security-focused systems like OpenBSD. As of April 2026, it is highly secure but not immune to AI-driven, sophisticated attacks.
https://en.wikipedia.org/wiki/List_of_products_based_on_FreeBSD
Trenutno je Claude Mythos on hold dok Project Glasswing ekipa ne penetrira svoje sustave/patchira i onda ga mozda opet puste u prodaju.
Ekipa iz Apple, Microsoft, Amazon (AWS), Google, NVIDIA, Cisco, and CrowdStrike se usrala kad su vidli kaj moze jedan AI.
Ta prica o mythosu je odavno prokazana kao myth. Malo istrazi...
