Data privacy in Large Language Models

From LemonWiki共筆
Jump to navigation Jump to search

Data privacy in Large Language Models


🌐 Switch language: English, 漢字


List of Large Language Models[edit]

Claude

  • Personal versions: User chat records may be used to improve AI models, but users can choose whether to participate in this training through privacy settings.[1][2]
  • Business versions: Conversation content is not used to train models[3]


OpenAI ChatGPT

  • Personal version: Personal version (non-Enterprise, non-Team, non-API plans) conversation content is used to train models [4][5]. Exceptions: (1) Enabling "Temporary Chat" will not use conversations to train OpenAI's models[6], (2) You can go to "OpenAI Privacy Center" to opt out of model training [7] Icon_exclaim.gif
  • Business version: Enterprise, Team, and API plan conversation content is not used to train models

Microsoft Copilot

  • Free version: Conversation content is used to train models[8]
  • Enterprise version (using Entra ID): No

Google Gemini

  • Enterprise and Education users: Conversation content is not used to train models and is not reviewed by human reviewers, with enterprise-grade data protection.[9]
  • Personal users: Conversation content may be used to train models and reviewed by humans. Do not share sensitive information.[10] Icon_exclaim.gif
  • Data retention periods:
    • Enterprise version: Admins can control conversation retention time (auto-delete after 3, 18, or 36 months, default 18 months)
    • Personal version: When activity recording is disabled, conversation content is retained in accounts for up to 72 hours[11]

Further reading[edit]

References[edit]