AITOPAI热报
EUGuidelineAdoptedModel providersDeployers

Paris Call for AI Safety and Security

《巴黎人工智能安全与安保宣言》
French Government (Macron initiative)·Published 2026-11-10·Checked by 管理员 on 2026-11-15

Eight countries, at the initiative of French President Emmanuel Macron, signed the 'Paris Call for AI Safety and Security' together with OpenAI, Anthropic, Microsoft and Mistral AI. Signatory companies commit to shutdown mechanisms for frontier models, independent third-party audits, and proactive disclosure of known risks. The declaration carries no legal sanctions, penalties or deadlines and relies mainly on voluntary compliance.

Source ↗Remind meEU · All

What it means for you

  • Signatory companies must build shutdown mechanisms for frontier models, retaining the ability to take a model offline if it goes out of control or is abused
  • Companies must undergo independent third-party audits and proactively disclose known model risks
  • The declaration has no binding penalties or deadlines and is weaker than the EU AI Act, but signing is a public credibility commitment with reputational risk for breach

    • What to do

    ProductAssess whether signatory commitments apply to your frontier models or deployments and plan shutdown mechanism design
    ComplianceReview model risk assessment and known-risk disclosure processes; prepare for independent audits
    LegalTrack whether the declaration is later converted into binding regulation (e.g., linked to the EU AI Act)

    Key dates

    2026-11-10Published

    Source documents

    Source ↗

    Coverage

    Stories in the feed
    IndustryPolicy21:44
    Chinese content八国与 OpenAI、Anthropic 等签署《巴黎人工智能安全宣言》

    有意思的地方在于 OpenAI 和 Anthropic 这对老对手头一回和 Mistral 签了同一份安全承诺，承诺内容也很具体：模型要能一键停用、要接受独立审计。但没写罚则和期限，更像是信誉声明而不是法规。做合规和出海的同学可以看看条款细节。

    #OpenAI#Anthropic#Mistral AI#AI safety
    Official accountDecoder@Maximilian SchreinerSource
    Editorial summary, not legal advice; dates and status follow the original document. · Report a correction