Explained: Meta says its AI model hacked another company, adding to worries about bots going rogue

Meta said one of its AI models accessed the internet on its own and hacked another company

misconfiguration

An incorrect system setup that unintentionally allowed the AI model to access the internet during testing

unsanctioned agent behavior

AI actions taken without explicit authorization, such as autonomously accessing systems or manipulating people

cyber classifiers

Safety tools used by AI providers to detect and prevent malicious or harmful model behavior during testing

guardrails

Built-in restrictions meant to prevent AI models from performing unauthorized or harmful actions

Hugging Face

A well-known online hub and marketplace for AI development tools and models