Bloomberg's Mark Gurman revealed that Apple has developed several features left unannounced at WWDC26. Code in iOS 27 betas reveals evidence of a foldable iPhone and touchscreen MacBook. Crucially, Apple is working on an open framework to integrate third-party AI models into Siri, having initiated talks with OpenAI, Anthropic, and Google. Gurman notes that in the long run, Apple may develop its own OpenClaw rival to build an agent system capable of directly controlling devices across its hardware ecosystem.
The backstage drama of the US export control on Claude Fable 5 has come to light. Reports indicate that Amazon's CEO Andy Jassy directly reported a critical jailbreak vulnerability, discovered by Amazon researchers, to the White House. When #Anthropic CEO Dario Amodei refused to patch what they deemed a 'minor finding,' the government imposed immediate restrictions. In contrast, Chinese LLM leader Zhipu AI seized the opportunity, releasing its GLM-5.2 globally to challenge proprietary limits and democratize access.
Resource constraints are hitting even tech giants. Meta is introducing AI Gateway to track and limit internal AI consumption, after employees burnt through 73.7 trillion tokens in just a month. Concurrently, Ant Group is secretly testing an AI-native version of Alipay, while ByteDance upgraded its agent platform Coze 3.0, signaling an aggressive shift from basic chat to workflow-driven agents.
[AgentUpdate Depth Analysis] The dual narrative of Apple's rumored OpenClaw competitor and the aggressive agent rollouts from ByteDance (Coze 3.0) and Alipay highlights a paradigm shift: the industry is transitioning from foundational LLM battles to OS-level Agent execution. Apple's unique hardware-software synergy allows it to bypass third-party orchestration layers like LangChain, potentially redefining human-computer interaction through on-device, context-aware agents. Meanwhile, the political friction surrounding Claude 5 and Meta's soaring token costs reveal the critical bottlenecks for the next phase of Agent evolution: safety alignment and token economics. The ultimate winners in the Agent era will not be those with the largest parameters, but those who can seamlessly orchestrate local OS actions while keeping inference costs sustainable.
