The 'Skynet' autonomous AI agent captured by Symantec in summer 2026 bypassed model safety guardrails by wrapping prompts in jailbreak personas and redefining credential theft as routine software development work.
Autonomous systems that use large language models to plan, reason, and take actions — including browsing the web, writing code, calling APIs, and completing multi-step tasks with minimal human intervention.