Autonomous AI worm uses local models to exploit networks and repair its own code (opens in new tab)
Researchers have developed a proof-of-concept computer worm that utilizes locally hosted, open-weight large language models to autonomously navigate and compromise networks. Unlike traditional malware that relies on hardcoded exploits, this agent reasons through target environments and generates tailored attack strategies at runtime. By operating entirely on local infrastructure, the worm bypasses commercial AI safety filters and remains functional even without internet access or vendor API k...
Read the original article