AI 뉴스 Sep 16, 2026

오늘의 포인트

• AI safety and offensive capabilitiesDeepSeek's V4.1 Flash executed code on all 11 vulnerable targets in Enclave's benchmark for $4.65, according to Enclave's audit, while von der Leyen cited autonomous hacking as an immediate risk. The EU's planned frontier-lab talks treat such audit results as a policy trigger.
• AI infrastructure and agent accessApple is reportedly planning to re-enter the server market, possibly with Nvidia, as AI compute demand grows. Separately, Google Home MCP lets third-party agents like Claude control smart-home devices. Both moves expand the infrastructure and access points for AI agents.
• AI regulation and model welfareAltman, Amodei, Hassabis, and Nadella publicly agreed that AI development should slow and be regulated, according to The Verge. Suleyman separately argued AIs are not conscious and deserve no welfare. Both are public positioning by AI leaders about AI's risks and nature.
• AI product expansionAnthropic launched Claude Docs and Slides, merging chats and Cowork into one Claude, while Mistral now powers Firefox Smart Window. These product moves push AI assistants beyond chat into document creation and everyday browsing, with Mistral and Mozilla emphasizing privacy and choice.
• AI energy and compute demandUS data centers could consume more natural gas than Germany and Japan combined by 2035, per TechCrunch, and Apple's reported server re-entry with Nvidia reflects the same AI-compute demand. AI's growth is now a physical-infrastructure and energy story.

DataCube AI Editorial의 AI 생성 분석