Taiwan's cybersecurity institute announces National Cybersecurity Protection 2.0; starting April 2027, AI agents will conduct red team penetration testing on government agencies.

Gong Huazhong, president of Taiwan's National Institute of Cyber Security, announced the "National Cybersecurity Protection 2.0" plan at a media exchange, with "using AI to fight AI" as the core strategy, advancing four key tasks. The first is AI-automated source code detection and repair, combining partner AI vendors' systems with the institute's self-built on-premises cybersecurity models, prioritizing the discovery of vulnerabilities and repair recommendations in core software of government agencies and critical infrastructure; this is currently in testing and validation. The second is using AI to strengthen penetration testing and attack-defense exercises, employing AI models, AI agents, and self-developed techniques to simulate attacker thinking, conducting automated black-box penetration testing and red-blue team attack-defense exercises on important online services and systems, expected to officially launch in April next year. The third is establishing an AI-automated threat intelligence system, introducing large language models to instantly digest and compare dark web and various countries' threat intelligence. The fourth is building a national-level AI cybersecurity attack-defense sandbox, providing a secure and controlled validation environment.