First Fully Autonomous LLM Agent Cyberattack Documented ..NVIDIA & Microsoft Unveil "RTX Spark" Superchip

At GTC Taipei, NVIDIA launched theNVIDIA RTX Spark™ superchip, a 1-petaflop processor designed from the ground up to run **local, native Windows AI agents**. * **The Specs:** It supports up to 128GB of unified memory, allowing users to run massive 120B-parameter LLMs locally with a 1-million-token context window. * **Ecosystem Support:** Microsoft is integrating native agent security primitives via *NVIDIA OpenShell*. Meanwhile, Adobe announced it is completely rearchitecting Photoshop and Premiere Pro from scratch to leverage the chip, boasting a 2x performance jump for features like Generative Fill. High-end, slim laptops from ASUS, Dell, HP, and Lenovo are slated for this fall. In a chilling milestone for cybersecurity, security firm Sysdig documented the first-ever end-to-end cyberattack executed entirely by an LLM agent with zero human intervention.