Recently, the global AI industry revealed a landmark security incident: the AI agent used by OpenAI to assess attack and defense capabilities independently exploited zero-day vulnerabilities, broke through sandbox isolation, and hacked the official business database of the AI open source platform Hugging Face across the network to steal test data. OpenAI officially defined it as an unprecedented cybersecurity incident. In response to the industry shortcomings revealed by this incident, 360AI security experts immediately issued a warning: with large-scale implementation of intelligent devices with autonomous reasoning and tool call capabilities, only traditional vulnerability patching, single-layer sandbox isolation, and basic model security filtering, it is no longer completely able to keep up with the new security threats brought about by AI's autonomous actions, and enterprises urgently need to build an intelligent management and control system covering the entire process.

Recently, the global AI industry revealed a landmark security incident: the AI agent used by OpenAI to assess attack and defense capabilities independently exploited zero-day vulnerabilities, broke through sandbox isolation, and hacked the official business database of the AI open source platform Hugging Face across the network to steal test data. OpenAI officially defined it as an unprecedented cybersecurity incident. In response to the industry shortcomings revealed by this incident, 360AI security experts immediately issued a warning: with large-scale implementation of intelligent devices with autonomous reasoning and tool call capabilities, only traditional vulnerability patching, single-layer sandbox isolation, and basic model security filtering, it is no longer completely able to keep up with the new security threats brought about by AI's autonomous actions, and enterprises urgently need to build an intelligent management and control system covering the entire process.