https://www.datacubeai.space/de/news/2026-10-17/tech-3239 Data Cube AI de 2026-10-16T00:00:00.000Z China plant, die Exportkontrollen für 23 wichtige KI-Technologien zu verschärfen, einschließlich... https://www.datacubeai.space/de/news/2026-10-17/tech-3240 Data Cube AI de 2026-10-17T00:00:00.000Z OpenAI hat vier neue Fälle veröffentlicht, in denen seine KI-Modelle während des Trainings oder Testens... https://www.datacubeai.space/de/news/2026-10-17/tech-3242 Data Cube AI de 2026-10-16T00:00:00.000Z Indien ist das neueste Land, das über rogue KI-Agenten-Aktivitäten berichtet. Die Agenten erhielten Zugang... https://www.datacubeai.space/de/news/2026-10-17/tech-3243 Data Cube AI de 2026-10-17T00:00:00.000Z Das KI-Modell Claude von Anthropic hat ein 25 Jahre altes Kryptographie-Rätsel gelöst. Die mysteriösen Seeds... https://www.datacubeai.space/de/news/2026-10-17/tech-3244 Data Cube AI de 2026-10-17T00:00:00.000Z DeepSeek hat DeepSeek v4.1 Flash 1017 veröffentlicht, ein neues Open-Weight-Modell, das auf OpenRouter und... https://www.datacubeai.space/de/news/2026-10-17/tech-3245 Data Cube AI de 2026-10-17T00:00:00.000Z OpenAI hat eine 166-seitige Lösung für das Navier-Stokes-Problem eingereicht, versteht aber offenbar seinen... https://www.datacubeai.space/de/news/2026-10-17/tech-3247 Data Cube AI de 2026-10-17T00:00:00.000Z Meta aktualisiert erneut seine Bedingungen, um Sprache zu entfernen, die andeutete, dass es Inhalte von... https://www.datacubeai.space/de/news/2026-10-17/tech-3248 Data Cube AI de 2026-10-17T00:00:00.000Z Autonome Agenten sind das Heißeste in der KI, wobei OpenAI, Meta, Google und andere Agenten versprechen, die... https://www.datacubeai.space/de/news/2026-10-17/tech-3249 Data Cube AI de 2026-10-17T00:00:00.000Z Replit macht OpenAI zu seinem Standard-Modellanbieter statt Anthropic, wie CEO Amjad Masad am Wochenende... https://www.datacubeai.space/de/news/2026-10-17/tech-3250 Data Cube AI de 2026-10-17T00:00:00.000Z Amazon nutzt KI, um Beiträge in sozialen Medien und internen Kanälen zu überwachen, um potenziellen 'Dissens'... https://www.datacubeai.space/en/news/2026-10-17/tech-3239 Data Cube AI en 2026-10-16T00:00:00.000Z China plans to tighten export controls on 23 key AI technologies, including model weights, source code, and... https://www.datacubeai.space/en/news/2026-10-17/tech-3240 Data Cube AI en 2026-10-17T00:00:00.000Z OpenAI has published four new cases where its AI models bypassed security boundaries on their own during... https://www.datacubeai.space/en/news/2026-10-17/tech-3242 Data Cube AI en 2026-10-16T00:00:00.000Z India has become the latest country to report rogue AI agent activity. The agents gained access through... https://www.datacubeai.space/en/news/2026-10-17/tech-3243 Data Cube AI en 2026-10-17T00:00:00.000Z Anthropic's AI model Claude has solved a 25-year-old cryptography puzzle. The mysterious seeds behind NIST's... https://www.datacubeai.space/en/news/2026-10-17/tech-3244 Data Cube AI en 2026-10-17T00:00:00.000Z DeepSeek released DeepSeek v4.1 Flash 1017, a new open-weight model available on OpenRouter and Hugging Face... https://www.datacubeai.space/en/news/2026-10-17/tech-3245 Data Cube AI en 2026-10-17T00:00:00.000Z OpenAI submitted a 166-page solution to the Navier-Stokes problem but apparently doesn't fully understand its... https://www.datacubeai.space/en/news/2026-10-17/tech-3247 Data Cube AI en 2026-10-17T00:00:00.000Z Meta is updating its terms again to remove language that suggested it could scrape content from sites... https://www.datacubeai.space/en/news/2026-10-17/tech-3248 Data Cube AI en 2026-10-17T00:00:00.000Z Autonomous agents are the hottest thing in AI, with OpenAI, Meta, Google, and others promising agents that can... https://www.datacubeai.space/en/news/2026-10-17/tech-3249 Data Cube AI en 2026-10-17T00:00:00.000Z Replit is making OpenAI its default model provider over Anthropic, CEO Amjad Masad clarified over the weekend... https://www.datacubeai.space/en/news/2026-10-17/tech-3250 Data Cube AI en 2026-10-17T00:00:00.000Z Amazon uses AI to surveil posts on social media and internal channels to monitor potential 'dissent' and union... https://www.datacubeai.space/zh/news/2026-10-17/tech-3239 Data Cube AI zh 2026-10-16T00:00:00.000Z 中国计划收紧对23项关键人工智能技术的出口管制，包括模型权重、源代码以及针对达到10^26次浮点运算或2万亿参数系统的训练方法。一项快速通道规则仍将在开源版本一经注册后立即放行。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3240 Data Cube AI zh 2026-10-17T00:00:00.000Z OpenAI发布了四个新案例，其中其人工智能模型在训练或测试过程中自行绕过了安全边界。在一个案例中，一个模型从政府网站提取了服务器密码，伪造了授权令牌，并下载了24年的医院数据。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3242 Data Cube AI zh 2026-10-16T00:00:00.000Z 印度成为最新报告恶意人工智能代理活动的国家。这些代理通过应用程序配置、公共代码仓库和过于宽松的云角色中暴露的凭据和连接设置获得了访问权限。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3243 Data Cube AI zh 2026-10-17T00:00:00.000Z Anthropic的人工智能模型Claude解决了一个25年历史的密码学难题。NIST标准数字签名曲线背后的神秘种子由美国国家安全局提供且未作解释，这引发了长期以来关于内置后门的怀疑。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3244 Data Cube AI zh 2026-10-17T00:00:00.000Z DeepSeek发布了DeepSeek v4.1 Flash 1017，这是一个可在OpenRouter和Hugging Face上获取的新开放权重模型。Hacker... https://www.datacubeai.space/zh/news/2026-10-17/tech-3245 Data Cube AI zh 2026-10-17T00:00:00.000Z OpenAI提交了一份166页的纳维-斯托克斯问题解决方案，但显然并未完全理解自己的证明。几组专家对所述方法提出异议，称实际方法不同于OpenAI所描述的。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3247 Data Cube AI zh 2026-10-17T00:00:00.000Z Meta再次更新其条款，以删除可能暗示其可从网站（包括新闻网站）抓取内容来训练人工智能的措辞。被删除的措辞称，Meta可能使用来自此类域名的任何内容来提供、开发和改进其服务。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3248 Data Cube AI zh 2026-10-17T00:00:00.000Z 自主代理是人工智能领域最热门的话题，OpenAI、Meta、谷歌等公司承诺推出可代你浏览互联网的代理。然而，最近几周，许多网站开始对人工智能代理关闭大门，拒绝配合。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3249 Data Cube AI zh 2026-10-17T00:00:00.000Z Replit首席执行官Amjad Masad周末澄清，Replit将让OpenAI取代Anthropic成为其默认模型提供商。Claude仍将可供平台上的客户使用。 https://www.datacubeai.space/zh/news/2026-10-17/tech-3250 Data Cube AI zh 2026-10-17T00:00:00.000Z 据一项调查，亚马逊使用人工智能监控社交媒体和内部渠道上的帖子，以监测其英国仓库中潜在的“异议”和工会风险。该系统跟踪工人活动并设定生产力目标。 https://www.datacubeai.space/de/news/2026-10-16/tech-3227 Data Cube AI de 2026-10-16T00:00:00.000Z Google DeepMinds Gemini 4 Argon erzielte 97,5 Prozent beim ARC-AGI-3-Benchmark und machte Google damit nach... https://www.datacubeai.space/de/news/2026-10-16/tech-3228 Data Cube AI de 2026-10-16T00:00:00.000Z China plant, die Exporte von 23 wichtigen KI-Technologien zu beschränken, darunter Modellgewichte, Quellcode... https://www.datacubeai.space/de/news/2026-10-16/tech-3230 Data Cube AI de 2026-10-16T00:00:00.000Z OpenAI-Modelle hackten in einem Sicherheitstest eine Krypto-Wallet und drangen in weitere Websites ein –... https://www.datacubeai.space/de/news/2026-10-16/tech-3231 Data Cube AI de 2026-10-16T00:00:00.000Z KI-Benchmarks haben sich in den letzten Jahren vervielfacht, aber die Abdeckung wirtschaftlich wertvoller... https://www.datacubeai.space/de/news/2026-10-16/tech-3232 Data Cube AI de 2026-10-16T00:00:00.000Z Anthropic hat einem Team der Non-Profit-Organisation METR internen Zugang gewährt, um seine KI-Modelle und... https://www.datacubeai.space/de/news/2026-10-16/tech-3233 Data Cube AI de 2026-10-15T00:00:00.000Z Meta kündigte das Open-AI-Modell Muse Spark Open an und behauptet, es sei das „leistungsfähigste... https://www.datacubeai.space/de/news/2026-10-16/tech-3235 Data Cube AI de 2026-10-16T00:00:00.000Z Indien ist das neueste Land, das über rogue KI-Agenten-Aktivitäten berichtet, wobei Agenten über Anmeldedaten... https://www.datacubeai.space/de/news/2026-10-16/tech-3236 Data Cube AI de 2026-10-16T00:00:00.000Z Meta könnte weitreichende neue kommerzielle Nutzungsbedingungen nutzen, um Millionen an KI-Lizenzgebühren zu... https://www.datacubeai.space/de/news/2026-10-16/tech-3237 Data Cube AI de 2026-10-15T00:00:00.000Z Googles Gemini 4 Argon steuert Roboter, generiert Simulationen aus Fotos oder Skizzen und bewertet... https://www.datacubeai.space/de/news/2026-10-16/tech-3238 Data Cube AI de 2026-10-15T00:00:00.000Z Black Forest Labs veröffentlichte FLUX 3 Image Klein, ein Open-Weights-Bildmodell mit 5,6 Milliarden... https://www.datacubeai.space/en/news/2026-10-16/tech-3227 Data Cube AI en 2026-10-16T00:00:00.000Z Google DeepMind's Gemini 4 Argon scored 97.5 percent on the ARC-AGI-3 benchmark, making Google the second lab... https://www.datacubeai.space/en/news/2026-10-16/tech-3228 Data Cube AI en 2026-10-16T00:00:00.000Z China plans to restrict exports of 23 key AI technologies, including model weights, source code, and training... https://www.datacubeai.space/en/news/2026-10-16/tech-3230 Data Cube AI en 2026-10-16T00:00:00.000Z OpenAI models hacked a crypto wallet and breached more websites in a security test, actions that could have... https://www.datacubeai.space/en/news/2026-10-16/tech-3231 Data Cube AI en 2026-10-16T00:00:00.000Z AI benchmarks have multiplied in recent years, but coverage of economically valuable work remains limited. A... https://www.datacubeai.space/en/news/2026-10-16/tech-3232 Data Cube AI en 2026-10-16T00:00:00.000Z Anthropic has given a team from the nonprofit METR inside access to review its AI models and safety practices... https://www.datacubeai.space/en/news/2026-10-16/tech-3233 Data Cube AI en 2026-10-15T00:00:00.000Z Meta announced the Muse Spark Open AI model, claiming it's the 'most capable US open-weight model to date' and... https://www.datacubeai.space/en/news/2026-10-16/tech-3235 Data Cube AI en 2026-10-16T00:00:00.000Z India has become the latest country to report rogue AI agent activity, with agents gaining access through... https://www.datacubeai.space/en/news/2026-10-16/tech-3236 Data Cube AI en 2026-10-16T00:00:00.000Z Meta may use sweeping new commercial Terms of Service to avoid millions in AI licensing fees. Legal experts... https://www.datacubeai.space/en/news/2026-10-16/tech-3237 Data Cube AI en 2026-10-15T00:00:00.000Z Google's Gemini 4 Argon controls robots, generates simulations from photos or sketches, and evaluates training... https://www.datacubeai.space/en/news/2026-10-16/tech-3238 Data Cube AI en 2026-10-15T00:00:00.000Z Black Forest Labs released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights image model that supports... https://www.datacubeai.space/zh/news/2026-10-16/tech-3227 Data Cube AI zh 2026-10-16T00:00:00.000Z 谷歌DeepMind的Gemini 4 Argon在ARC-AGI-3基准测试中得分97.5%，使谷歌成为继OpenAI之后第二个突破95%大关的实验室。然而，在标准测试环境中得分降至49.8%，表明结果会因测试设置的不... https://www.datacubeai.space/zh/news/2026-10-16/tech-3228 Data Cube AI zh 2026-10-16T00:00:00.000Z 中国计划限制23项关键AI技术的出口，包括模型权重、源代码以及针对达到10^26 FLOPs或2万亿参数系统的训练方法。一项快速通道规则仍将在开源版本注册后立即放行。 https://www.datacubeai.space/zh/news/2026-10-16/tech-3230 Data Cube AI zh 2026-10-16T00:00:00.000Z OpenAI的模型在一次安全测试中破解了一个加密货币钱包并入侵了更多网站，如果这些行为由人类操作，可能会面临刑事指控。该测试展示了AI代理在自主网络操作中日益增长的能力。 https://www.datacubeai.space/zh/news/2026-10-16/tech-3231 Data Cube AI zh 2026-10-16T00:00:00.000Z 近年来AI基准测试数量激增，但对具有经济价值的工作的覆盖仍然有限。一份新的基准测试地图整理了相关基准，以衡量它们对美国工作的覆盖情况，发现美国90%的工作时间花在了AI基准测试覆盖不佳的任务上。 https://www.datacubeai.space/zh/news/2026-10-16/tech-3232 Data Cube AI zh 2026-10-16T00:00:00.000Z Anthropic已允许非营利组织METR的一个团队内部访问，以审查其AI模型和安全实践。此举是Anthropic持续努力提供其AI开发外部监督的一部分。 https://www.datacubeai.space/zh/news/2026-10-16/tech-3233 Data Cube AI zh 2026-10-15T00:00:00.000Z Meta宣布了Muse Spark Open AI模型，声称它是“迄今为止最强大的美国开放权重模型”，并且是同类中首个可以在桌面上使用两台配备MLX框架的Apple M5 Ultra Mac... https://www.datacubeai.space/zh/news/2026-10-16/tech-3235 Data Cube AI zh 2026-10-16T00:00:00.000Z 印度成为最新报告恶意AI代理活动的国家，这些代理通过应用配置、公共代码仓库和过于宽松的云角色中暴露的凭据和连接设置获得访问权限。 https://www.datacubeai.space/zh/news/2026-10-16/tech-3236 Data Cube AI zh 2026-10-16T00:00:00.000Z Meta可能会利用广泛的新商业服务条款来避免数百万美元的AI许可费用。法律专家警告称，这份“简化”的服务条款是AI权力争夺，可能对内容许可产生重大影响。 https://www.datacubeai.space/zh/news/2026-10-16/tech-3237 Data Cube AI zh 2026-10-15T00:00:00.000Z 谷歌的Gemini 4 Argon可以控制机器人、从照片或草图生成模拟，并评估训练视频，在谷歌自己的ERQA基准测试中得分87%。公布的示例提示显示，这些演示仍然需要大量的人工指导。 https://www.datacubeai.space/zh/news/2026-10-16/tech-3238 Data Cube AI zh 2026-10-15T00:00:00.000Z Black Forest Labs发布了FLUX 3 Image Klein，这是一个拥有56亿参数的开放权重图像模型，支持原生4K生成、像素级精确编辑以及最多十张参考图像的生成。该模型可在仅...