來源GitHub Blog•較早收集於 20m
GitHub Copilot 互動資料使用政策更新

#data-privacy#ai-training#opt-out#policy-changegithub-copilotgithubcopilot
💡Copilot 將從4月24日起使用你的程式碼資料訓練,除非退出—立即保護智慧財產!
⚡ 30 秒速覽
有什麼變化
從4月24日起,Copilot 資料用於 AI 訓練,除非選擇退出
為什麼重要
這讓 GitHub 能以真實使用者資料精進 Copilot,提升程式設計師的 AI 品質。但可能洩露專有程式碼,開發者應為隱私選擇退出。
下一步行動
在4月24日前,前往 GitHub Copilot 設定並選擇退出互動資料使用。
誰應關注:Developers & AI Engineers
關鍵要點
- •從4月24日起,Copilot 資料用於 AI 訓練,除非選擇退出
- •適用於 Free、Pro、Pro+ 方案
- •資料包括輸入、輸出、程式碼片段及上下文
🧠 深度解析
本篇為 AI 生成分析,非原文內容。
🔑 增強重點摘要
- •GitHub has introduced a centralized 'Privacy Center' dashboard to manage these opt-out settings, moving away from previous granular configuration methods.
- •The policy update explicitly excludes Enterprise-tier customers, who retain default 'zero-retention' policies for their telemetry and code data.
- •This shift aligns GitHub's data usage policy with broader Microsoft 'Responsible AI' initiatives, aiming to reduce model hallucinations by incorporating real-world, non-proprietary developer patterns.
📊 競品分析▸ Show
| Feature | GitHub Copilot | Cursor | Tabnine |
|---|---|---|---|
| Model Training | Opt-out (User data) | Opt-in (User data) | Local/Private options |
| Pricing | Free/Pro/Pro+/Ent | Free/Pro/Business | Free/Pro/Enterprise |
| Privacy Focus | Enterprise-grade | Privacy-first/Local | Air-gapped/On-prem |
🛠️ 技術深入
- •The training pipeline utilizes a 'de-identification' layer that attempts to scrub PII (Personally Identifiable Information) and secrets from code snippets before ingestion into the training corpus.
- •Data is processed using a differential privacy mechanism to ensure that individual code patterns cannot be reconstructed or reverse-engineered from the resulting model weights.
- •The telemetry pipeline uses a 'sliding window' approach to capture context, which includes the active file, surrounding open tabs, and recent git commit history to improve model grounding.
🔮 前景展望基於引用來源的 AI 分析
Increased enterprise adoption of 'Zero-Data' tiers.
Organizations concerned about intellectual property leakage will likely migrate to Enterprise tiers to avoid the default data-sharing policy of lower-cost plans.
Regulatory scrutiny regarding 'implied consent' for data training.
The shift to an opt-out model rather than opt-in may trigger investigations from data privacy regulators in jurisdictions like the EU under GDPR.
⏳ 時間線
2021-06
GitHub Copilot technical preview launched.
2022-06
GitHub Copilot becomes generally available for individual developers.
2023-03
GitHub Copilot X announced, integrating GPT-4 and chat capabilities.
2024-05
GitHub introduces Copilot Extensions to allow third-party tool integration.
2025-11
GitHub updates Copilot to support multi-modal reasoning across entire repositories.
📰
AI 週報
閱讀本週精選 AI 大事摘要 →
👉相關動態
AI 策展新聞聚合。所有內容版權歸原始發布者所有。
原始來源: GitHub Blog ↗
每週電子報
每週一封,可隨時退訂。