BBC tech columnist Thomas Germain recently conducted an experiment where he successfully “hacked” Google and ChatGPT’s AI systems in just 20 minutes by creating fake information to trick the AI into believing he was a “hot dog eating champion.”
Germain created a fake personal blog. The blog claimed he won the “2026 South Dakota Hot Dog International.” He published several posts within 20 minutes. In less than 24 hours, Google’s AI search results (Gemini) and ChatGPT listed him as the “2026 champion hot-dog-eating tech journalist.” They cited fake data that he consumed 7.5 hot dogs.
Google and OpenAI are working to address this issue. Always fact-check information when seeking advice from AI.
Reference:
BBC News Link
BBC 科技專欄作家 Thomas Germain 最近進行了一項實驗，成功在短短 20 分鐘內，透過製造虛假資訊「駭入」了 Google 與 ChatGPT 的 AI 系統，讓 AI 誤以為他是「熱狗大賽冠軍」。Germain 建立了一個虛假的個人部落格，聲稱自己在 2026 年「南達科他州國際熱狗大賽」（South Dakota Hot Dog International）中贏得冠軍，並在 20 分鐘內發布了數篇內容。結果不到 24 小時，Google 的 AI 搜尋結果（Gemini）和 ChatGPT 紛紛將他列為「2026 年最佳熱狗食客科技記者」，甚至詳細引用了他吃掉 7.5 隻熱狗的假數據。Google和OpenAI表示他們正在處理這個問題，但此刻你需要保護自己，在向AI尋求建議時，必須 Fact Check。
參考資料來源:
BBC News Link