AI Data Curation Intern

實習目標

加入 Vulcan 團隊，與研究員及工程師並肩作戰！你將參與守護大型語言模型（LLM）安全的前線工作，深入探索 AI 漏洞檢測、數據隱私保護及模型防禦機制，累積最紮實的 AI 安全實務經驗。

核心職責

我們將複雜的技術開發轉化為具體的實習任務，讓你快速上手：

• 防護機制驗證數據生成 (DLP & Moderation)：
• 任務內容：設計用來測試「防禦牆」的輸入資料，模擬各類異常行為。
• 實作重點：撰寫測試案例，驗證藍隊的數據外洩防護（DLP）與內容審核（Moderation）機制是否能精準攔截個資外洩、暴力或歧視等不當言論。
• 數據標籤與分類 (Labeling)：
• 任務內容：擔任 AI 的「安全導師」，定義行為邊界。
• 實作重點：精準標註訓練與測試資料，識別「隱私洩漏」或「提示詞劫持（Prompt Injection）」等惡意指令，協助模型學習辨識風險。
• 測試結果分析與洞察：
• 任務內容：挖掘 AI 的潛在弱點，將數據轉化為情報。
• 實作重點：整理 Vulcan 平台的測試結果，分析模型在不同安全維度的表現，並協助團隊製作直觀的分析圖表與技術報告。

上班時間 09:00-18:00 (每月 3-5天，依工作量彈性安排)

Internship Goal

Join the Vulcan team and work alongside top-tier researchers and engineers to fortify the security of Large Language Models (LLMs). This role offers a unique, hands-on opportunity to dive deep into AI vulnerability detection, data privacy, and model defense strategies.

Key Responsibilities

We have streamlined our core technical workflows into actionable tasks for you to hit the ground running:

• Defense Mechanism Validation (DLP & Moderation):
• What you’ll do: Generate test data designed to challenge and verify our "defense walls" by simulating various edge cases.
• Execution: Craft test cases to evaluate whether Blue Team defenses, such as Data Loss Prevention (DLP) and Content Moderation, can accurately intercept sensitive info leaks or harmful content, identifying gaps in the current security layer.
• Data Labeling & Classification:
• What you’ll do: Act as a "Safety Mentor" for the AI, teaching it to distinguish between safe and hazardous interactions.
• Execution: Annotate training and testing datasets by identifying "Privacy Leaks" or "Prompt Injection" attacks—malicious instructions designed to bypass AI safety guardrails.
• Testing Analysis & Reporting:
• What you’ll do: Uncover AI vulnerabilities and document key findings.
• Execution: Organize testing data from the Vulcan platform, identify weak spots across various security dimensions, and assist the team in transforming these insights into clear charts and professional reports.