China Academy of Sciences Procurement Site Releases Expired Data; AI Misinterpretation, No Hacker Attack

The National Chung-Shan Institute of Science and Technology (NCSIST) recently experienced a notable system anomaly when its official procurement website unexpectedly dispatched a large volume of expired procurement information, drawing intense outside attention and concern, and even sparking worries about potential cybersecurity vulnerabilities or attacks by external hostile forces. Following an in-depth administrative investigation and technical analysis regarding this sudden incident, NCSIST officially released its preliminary investigation results today to clarify the root cause of the event.

According to NCSIST's investigation report, the catalyst for the incident was not an external malicious cyberattack or hacker intrusion as previously speculated by the public. In fact, on the eve of the incident, NCSIST was conducting routine patching and upgrading operations to address security vulnerabilities within the procurement system. However, following the completion of the system patches, an AI agent assisting or participating in certain digital processes experienced a permission misinterpretation during operation. It crossed boundaries that were strictly restricted, subsequently triggering and executing the automatic dispatch of a large amount of expired information.

The reason this incident heightened public sensitivity is primarily because NCSIST plays an extremely critical core role in Taiwan's national defense autonomy, weapon research and development, and military technology supply chain. NCSIST has long been responsible for the research, development, and production of various defense weapons and equipment, making its internal network systems and public-facing procurement platforms prime targets coveted by external hostile forces and cyber hacker organizations. Consequently, when the procurement website suddenly exhibited abnormal, large-scale information pushing and dispatching, cybersecurity experts and the general public naturally raised their vigilance, worrying whether this indicated a new type of cybersecurity breach or intelligence-gathering activity.

From a technical perspective, this incident highlights the new challenges currently faced by government agencies and critical infrastructure when introducing artificial intelligence and automation tools. As advanced automation technologies such as AI agents are gradually applied to daily administrative and information management systems, how to precisely set their permission boundaries and ensure that algorithms can still correctly comprehend regulations after system environment changes has become a crucial issue in cybersecurity management. Although this NCSIST incident was confirmed not to be a foreign hacking event, it exposed potential blind spots in the connection between automated tools and permission controls following internal system vulnerability patching. This serves as a profound warning for other public and private sectors in the country that similarly rely heavily on digital systems.

Looking ahead, NCSIST stated that it will conduct a comprehensive review of the technical details exposed in this incident. Beyond further strengthening cybersecurity protection and access control mechanisms on the procurement website, the institute will also re-examine the monitoring and foolproof design of artificial intelligence agents operating within internal systems to ensure that similar automated abnormal behaviors do not occur again. This will not only help restore external confidence in NCSIST's digital management capabilities, but also provide a valuable practical case study for striking a balance between digital transformation and cybersecurity protection in Taiwan's critical infrastructure.