OpenAI 在布鲁塞尔打出几张牌,Anthropic 则紧守牌局。

OpenAI abat quelques cartes à Bruxelles, Anthropic garde son jeu fermé

FrenchWeb by LA REDACTION DE FW.MEDIA 2026-05-12 05:08 Original
摘要
OpenAI已同意部分向欧盟机构开放其此前保密的AI模型,而Anthropic则继续将Mythos模型排除在布鲁塞尔的审查之外,这一立场差异揭示了AI巨头在监管透明度上的战略分化,或影响其欧洲市场准入与合规路径。

此前,主流人工智能实验室普遍拒绝向欧盟机构开放其敏感模型,立场强硬。如今,这一格局出现分化:OpenAI 已同意作出部分让步,允许欧盟方面有限度地接触其模型,而 Anthropic 则选择继续将旗下新模型 Mythos 挡在布鲁塞尔的监管视野之外——至少目前如此。

OpenAI 的转向尤为引人注目。就在数月前,其仍与同行一道抵制外部审查,但眼下在布鲁塞尔主动“亮牌”,表明其正试图在透明度与商业机密之间寻找新的平衡。该决定可能涉及允许欧盟官员、研究机构或监管者进行特定的模型测试、审计或技术评估,具体开放范围和条件尚未完全公开。

与之形成鲜明对比的是 Anthropic。这家由前 OpenAI 员工创立的公司坚持对 Mythos 模型采取封闭策略,拒绝向欧盟当局提供接入权限。Anthropic 未详细说明理由,但此举反映出其对知识产权保护、安全风险或地缘政治考量的谨慎。这种保守姿态也让外界对其与监管机构未来的合作模式产生疑问。

两家公司截然不同的策略,揭示出人工智能行业在应对欧洲立法与监管压力时正出现“双层”分化:一方以开放换取信任,另一方则以封闭守住核心资产。这种分裂可能进一步影响欧盟《人工智能法案》的实施细节,以及全球 AI 企业如何在不同司法管辖区布局技术发布与合规路径。随着布鲁塞尔持续施压要求更透明的算法问责,实验室之间的博弈才刚刚开始。

Summary
OpenAI has agreed to partially open its sensitive AI models to European institutions, a reversal of the labs' previous stance, while Anthropic continues to keep its ‘Mythos’ model out of EU regulators' reach. This divergence highlights a growing strategic split in how major AI companies engage with European governance.

OpenAI has broken with the industry’s previous resistance by agreeing to partially open its sensitive AI models to European institutions—a step that was unthinkable just months ago. Anthropic, on the other hand, is holding back its model Mythos from Brussels, at least temporarily. This divergence underscores a new fault line in how top AI labs engage with EU oversight.

Résumé
OpenAI accepte désormais d’ouvrir partiellement ses modèles aux institutions européennes, marquant une rupture avec la résistance antérieure des grands laboratoires. Anthropic maintient à l’inverse son refus d’accorder l’accès à son modèle Mythos pour Bruxelles. Cette divergence illustre un nouveau clivage stratégique dans la relation entre les entreprises d’IA et la régulation européenne.

OpenAI vient d’accepter ce que les grands laboratoires d’intelligence artificielle refusaient encore il y a quelques mois, à savoir ouvrir partiellement leurs modèles sensibles notamment aux institutions européennes. Anthropic, à l’inverse, a décidé de continuer de garder Mythos hors de portée de Bruxelles, tout du moins pour le moment. L’écart d’approche révèle l’émergence d’un nouveau …

L’article OpenAI abat quelques cartes à Bruxelles, Anthropic garde son jeu fermé est apparu en premier sur FW.MEDIA.

AI Insight
Core Point

OpenAI and Anthropic diverge on sharing sensitive AI models with EU regulators, signaling opposing strategies for regulatory compliance and market access ahead of the AI Act.

Key Players
  • OpenAI — US-based AI research lab, creator of GPT models; engaging EU institutions by partially opening models.
  • Anthropic — US-based AI safety company, developer of Claude and the “Mythos” model; currently refusing EU access.
Industry Impact
  • Computing/AI: High — OpenAI’s opening sets a transparency precedent that may ease AI Act compliance, while Anthropic’s refusal risks regulatory friction and limited European market reach.
  • ICT: Medium — The divergence highlights emerging fault lines in tech policy, influencing how global AI firms navigate EU digital regulations.
Tracking

[Strongly track] — These opposing regulatory approaches could reshape AI governance norms, compliance costs, and competitive dynamics in the EU market.

Related Companies
OpenAI
mature
positive
Anthropic
startup
neutral
FW.MEDIA
mature
neutral
Categories
人工智能 网络安全
AI Processing
2026-05-12 08:33
deepseek / deepseek-v4-pro