
經濟部智慧局示警,民眾與企業員工使用AI工具時可能忽視資訊隱私,將營業秘密輸入如ChatGPT等生成式AI,恐在AI回答其他使用者問題時遭到外洩,衝擊企業競爭優勢。
AI-generated summary
經濟部智慧局針對人工智慧發展與企業資訊安全提出警告,指出生成式AI訓練機制可能導致機密外流。
經濟部智慧局示警AI洩密風險。(路透)
〔記者廖家寧/台北報導〕民眾依賴AI(人工智慧)方便性的同時可能忽視對資訊隱私的保護,AI正透過各界提供的大量對話資料訓練並強化自身功能,在此過程中伴隨資訊隱私與營業秘密外洩的風險。經濟部智慧局特別示警,就有案例是A公司員工將公司營業秘密輸入ChatGPT,B公司員工詢問ChatGPT類似問題,AI系統可能會使用其獲得的A公司員工所輸入涉及營業秘密內容來產生答案答覆B公司員工。
智慧局指出,企業的營業秘密被喻為「皇冠上的寶石」,企業投入大量研發成本的技術或Know-how,具有產業競爭優勢,是企業的重要資產,一旦外流,企業研發心血付諸流水,更面臨被競爭對手技術超車。
智慧局指出,常見的洩密樣態包括員工以電子郵件、隨身碟、電腦螢幕截圖或以手機拍照等方式重製營業秘密檔案,或以紙本影印攜出或提供新任職公司使用,或是員工在公務電腦安裝未獲公司授權的軟體,以該軟體下載公司的營業秘密檔案,上傳至個人硬碟後攜出公司使用等,或是協力廠商人員利用進入其他公司維護設備的機會,抄錄機器之製程參數等。
除常見的洩密樣態,智慧局特別示警AI洩密事件,從2023年4月南韓三星公司使用生成式AI 洩密事件的啟示中來看,公司員工A將公司的一項營業秘密輸入ChatGPT,該模型可能會從輸入中學習,如果有其他公司員工B向ChatGPT提出類似問題,AI系統可能會使用其了解的員工A所輸入涉及營業秘密內容來產生答案,從而使公司面臨被洩密的風險。

Lin Junxiu, director of the Digital Industry Agency of Taiwan's Ministry of Digital Development, led a delegation to the Netherlands to facilitate the signing of a chip security MOU between the Information Policy Council and the Dutch TrustCB, promote the integration of Taiwan's chip security standards with international certification, and communicate with organizations such as ASML, NXP and TNO.

Intel CEO Chen Liwu and Broadcom CEO Chen Fuyang shared the stage at the DBS Asia Foresight Summit in Singapore. Chen Liwu warned that the supply of AI computing power will continue to exceed demand until 2029, and pointed out power shortages and high capital costs as the main expansion bottlenecks; Chen Fuyang analyzed that customized accelerators will become the key, and the core value of AI is shifting to the application side.

New un-peer-reviewed research reveals a potential flaw in DeepSeek's memory-saving technology, showing models may miss information when it appears in certain document positions.

The new national standard GB/T 47746-2026 has been implemented, requiring the manual entrance to be clear and eye-catching, the human-machine switching to be smooth, and the enterprise to bear the main responsibility for the AI response. However, the reporter actually tested 20 mainstream platforms and mini programs and found that most platforms still need to enter "switch to manual" multiple times before they can access human customer service, and there are even cases where intelligent customer service pretends to be human. Industry insiders pointed out that this is not a technical problem, but a strategic obstacle caused by the company's inclusion of labor conversion rate as an efficiency assessment indicator. Experts suggest that through a mechanism that combines credit files, administrative supervision, and judicial adjudication, companies can be encouraged to optimize their customer service structures and return manual customer service to high-value service links.

China's manufacturing sector is skipping traditional digitalization and deploying AI agents directly, according to Black Lake CEO Zhou Yuxiang, aligning with Beijing's 'AI plus' initiative to integrate AI into 90% of the economy by 2030.

Hong Kong's Chief Secretary Eric Chan Kwok-ki said the government's AI city brain system, designed to integrate traffic, flood monitoring, drainage, weather and public inquiry data, could be tested as early as next typhoon season.