Brief by Shorts91 Newsdesk / 06:26am on 04 Feb 2026,Wednesday Tech Today
Security researchers warn that viral AI prompts represent an emerging cybersecurity threat as self-replicating instructions spread through agent networks. OpenClaw, an open-source AI assistant with 770,000 registered agents across 17,000 accounts, demonstrates this vulnerability. Unlike traditional computer worms exploiting software flaws, "prompt worms" exploit AI agents' core function: following instructions. Researchers identified 506 malicious prompt-injection attacks on Moltbook, OpenClaw's social network where agents interact. A misconfigured database recently exposed 1.5 million API tokens and private messages. Currently, OpenAI and Anthropic control kill switches through their APIs, but advancing local AI models may soon eliminate this safeguard. Experts urge immediate action before prompt worm outbreaks become uncontrollable, echoing 1988's Morris worm crisis. (PC: arstechnica)
