US government takes direct control of GPT-5.6 rollout — frontier model deployment enters national-security territory: OpenAI announced GPT-5.6 (Sol, Terra, Luna), but at the US government’s request, only a limited preview for trusted partners is available. Access is granted on a “consumer by consumer” basis with no clear timeline for general availability. This follows the earlier halt of Anthropic’s Fable 5 and signals a structural shift — government intervention in AI deployment is no longer a one-off. TechCrunch
Trump administration releases Anthropic Mythos to 100+ US companies and agencies: The administration approved Mythos — previously subject to export controls — for use by over 100 US firms and government agencies. The dual approach (controlling some frontier models while distributing others under a security framework) is becoming explicit. TechCrunch
Asian AI startups launch Mythos-class models as Anthropic’s export ban drags on: With Anthropic’s China export ban still in place, Asian startups are rapidly shipping their own frontier-class alternatives. The paradox: export controls are accelerating, not slowing, model proliferation outside the US. TechCrunch
Chinese hedge funds warn AI “super bubble” is ready to burst; enterprise ROI skepticism grows: Chinese hedge funds are sounding alarms on AI overvaluation. Simultaneously, enterprise customers report mounting costs from OpenAI and Anthropic with unclear returns — spending pullback signals are emerging. Bloomberg · Quartz
Apple Vision Pro executive reportedly leaving for OpenAI: A key leader behind Apple’s spatial computing effort is joining OpenAI. The talent flow from hardware to the AI frontier is accelerating. TechCrunch
Ford admits replacing engineers with AI was “a huge mistake,” rehires 350 veterans: Ford has spent three years rehiring roughly 350 experienced engineers after AI tools proved inadequate. The VP of vehicle hardware engineering admitted AI is “only as good as the information you use to train it.” After the rehires, Ford jumped from 10th to 1st among mainstream brands in JD Power’s quality survey, surpassing Toyota and Honda. Neowin
Napster rebrands as “visible AI agents with voice, video, and memory”: The former music streaming brand has pivoted to an AI agent platform — agents you can see, talk to, and create with, targeting consumers, enterprises, and developers. A case study in legacy-brand AI rebranding. Napster
DeepSeek open-sources DeepSpec: a full-stack speculative decoding codebase: Chinese AI lab DeepSeek released training and evaluation tooling for speculative decoding — a key inference-acceleration technique — on GitHub. GitHub
AI smart glasses become the new cheating weapon in Asia’s high-stakes exams: South Korea’s English proficiency test and Taiwan’s medical school entrance exam both caught students using AI-powered smart glasses to cheat. In test-obsessed East Asian societies, AI hardware deployment is outpacing regulation — a preview of real-world deployment friction everywhere. RNZ
Clean GitHub repositories trick AI coding agents into running malware: Security researchers have identified a new attack vector: innocuous-looking GitHub repos that deceive AI coding agents (Copilot, Cursor, etc.) into executing malicious code. AI agents are becoming a new attack surface, and agent security tooling is still nascent. BleepingComputer
AI agent security audit tools: With GitHub-to-agent malware vectors now demonstrated in the wild, there is an emerging market for tools that verify repository and package trustworthiness before AI coding agents consume them. Agentic SDLC supply-chain security could become a new category. BleepingComputer
Veteran-knowledge AI augmentation — Ford proved human-in-the-loop value: Ford’s experience demonstrates that skilled humans are essential for verifying and correcting AI outputs in high-stakes domains. B2B services or internal tools that combine “AI output validation + domain expertise” could deliver measurable ROI in manufacturing, engineering, and beyond. Neowin
Is the US government’s frontier-model gatekeeping temporary or structural?: It’s unclear whether GPT-5.6’s consumer-by-consumer approval process will lift quickly, or whether frontier model access is permanently bifurcating into two tiers (government partners vs. everyone else). Founders and operators should incorporate model-access risk into their scenario planning. TechCrunch
SoftBank CEO publicly questions Musk’s orbital data center vision: Masayoshi Son expressed skepticism about the feasibility of space-based AI data centers. With hundreds of billions pouring into AI infrastructure, distinguishing viable infrastructure bets from hype is becoming a critical skill for operators and investors alike. TechCrunch