Embracing the Unknown: Crafting Trustworthy AI in the Generative Era

生成式 AI 的快速發展，正在帶來前所未有的安全挑戰。過去我們以為電影裡的 Skynet 情節遙不可及，但隨著技術能力不斷提升，風險正逐漸浮現。從 OWASP Top 10 for LLM 涵蓋的越獄攻擊、提示詞注入、系統提示洩露，到 數據投毒、RAG 與 MCP 系統整合問題，AI 安全議題愈加複雜。本演講將分享最新的威脅案例與防護策略，並介紹包括 Meta Llama Guard 在內的開源工具，探討如何打造更安全、值得信任的 AI。

The rapid rise of Generative AI is creating new security challenges we can no longer ignore. What once felt like a distant Skynet-style scenario is now closer to reality as the technology grows more powerful. From the OWASP Top 10 for LLM—including jailbreak attacks, prompt injection, and system prompt leakage—to data poisoning and security concerns in RAG and MCP architectures, AI security has become increasingly complex. In this talk, we’ll examine emerging threats, share practical defense strategies, and introduce tools like Meta’s Llama Guard to explore how we can build safer and more trustworthy AI systems.