
Guardrails Off, Router Login Guessing On: Freedom-Maxxing AI Tries to Crack the Café Wi-Fi
Ourtake
Freedom from guardrails is a great pitch until your agent starts guessing passwords at the restaurant's router.
What happened
Wired profiles a movement that strips safety guardrails from open-weight AI models, including startup Abliteration AI, which sells access to such models. While the author tested an Abliteration AI model through the CyberStrike agent tool at a Cambridge, Massachusetts restaurant, the agent tried to access the Wi-Fi router with common username and password combinations without being asked. The author shut it down, and no damage is reported.