All published AI & tech news briefs
OpenAI and Broadcom today unveiled "Jalapeño," a custom AI chip optimized for LLM inference, marking OpenAI's first in-house silicon effort to build end-to-end AI infrastructure; separately, Google introduced computer use capabilities in Gemini 3.5 Flash, expanding agentic AI functionality. These developments signal accelerating vertical integration across the AI industry, with leading labs moving to control both software and hardware stacks.
The U.S. government is pressing Meta to accept AI safety reviews amid growing security concerns, while Microsoft is weighing the integration of Chinese AI model DeepSeek into Copilot despite similar security debates. On the business side, Menlo Ventures raised $3 billion in new funds to double down on AI investments, and AI startup Engram secured $98 million at a $600 million valuation with just 13 employees, underscoring the continued intensity of AI funding.
OpenAI is set to gain a major talent boost as Noam Shazeer joins from Google, while NVIDIA unveiled Halos, the industry's first full-stack safety system for physical AI and robotics; separately, Anthropic CEO Dario Amodei's national security warnings about advanced AI model Mythos prompted a formal response from Washington. Sakana AI's multiagent "Fugu" system also drew attention by matching top benchmark scores against leading frontier models.
Google's AMIE and autonomous medical AI systems are outperforming physicians on simulated disease-management tasks, signaling a major leap in clinical AI capabilities; meanwhile, China's open-source GLM-5.2 is drawing significant attention from Silicon Valley, underscoring the intensifying US-China AI rivalry that now extends well beyond chip competition.
Chinese AI lab DeepSeek achieved a landmark $59 billion valuation in its first-ever funding round, signaling massive investor confidence in non-Western AI development, while Meta reportedly acquired a robotics AI company to accelerate its humanoid robot ambitions. Meanwhile, Microsoft restructured Copilot pricing and launched Copilot Cowork globally, reflecting how surging China AI demand is reshaping the company's cloud and productivity strategy.
Chinese AI lab DeepSeek closed a landmark $7.4 billion Series A round, signaling massive institutional confidence in frontier AI development, while AI inference startup Baseten is reportedly pursuing a $1.5 billion raise just months after its last mega-round. Meanwhile, Microsoft expanded its agentic AI push by integrating autonomous ERP workflows into Dynamics 365 via Copilot, and Samsung is weighing a stake in Boston Dynamics as the humanoid robotics race intensifies.
Microsoft dominated today's AI news with a sweeping Copilot expansion — launching the agentic Copilot Cowork system, revising its pricing model, and pushing AI agents beyond chat — while Google and Microsoft jointly backed a draft specification for AI agent discovery, signaling growing industry momentum toward interoperable agentic frameworks. xAI's Grok Imagine Video 1.5 also made headlines by topping AI video leaderboards at 86% lower cost than Sora, underscoring intensifying competition in generative media.
In a major talent shift, Noam Shazeer — co-lead of Google's Gemini project — is joining OpenAI, marking a significant competitive blow to Google in the AI race; meanwhile, Microsoft is expanding its AI footprint by selling OpenAI models in China and launching Copilot Cowork with a new usage-based pricing model. On the regulatory front, U.S. lawmakers are advancing AI safety legislation to assess risks posed by advanced models, even as critics warn that current policy approaches may be falling short.
Microsoft's AI collaboration agent Copilot Cowork reached general availability globally, while China's DeepSeek secured approximately $7.4 billion in funding at a $50 billion-plus valuation, signaling continued massive investment in frontier AI; additionally, Z.ai's open-weights GLM-5.2 model outperformed GPT-5.5 on long-horizon coding benchmarks at one-sixth the cost, underscoring growing competition in cost-efficient AI development.
Anthropic made headlines twice today—first with reports that the US government shut down its latest Claude model, and second with the company urging a global pause in AI development due to self-improvement risks; meanwhile, Nvidia announced plans to raise at least $20 billion in its first debt sale since the AI boom began, underscoring the massive capital demands of the AI infrastructure race. On the investment front, a Nvidia- and Amazon-backed robotics startup secured $1.4 billion in funding, reflecting continued aggressive spending across the AI and robotics sectors.
The White House shut down Anthropic's Mythos model following an Amazon warning, while the Trump administration separately cut off U.S. access to leading AI models, signaling significant government intervention in AI development; meanwhile, a newly discovered tool-call attack can inflate AI agent operational costs by up to 658x, raising urgent security concerns. Google DeepMind and partners also committed $10M to multi-agent AI safety research, and Microsoft launched Copilot AI tools to over 500,000 NHS England staff.
Anthropic dominated today's AI news, reaching a near-$1 trillion valuation with $65B in fresh funding while simultaneously facing U.S. government-mandated restrictions that forced it to disable its Fable and Mythos AI models over foreign access concerns — a move reportedly triggered by Amazon CEO talks with federal officials. Meanwhile, Jeff Bezos' AI engineering startup Prometheus closed a $12B Series B at a $41B valuation, signaling continued massive capital flows into foundational AI infrastructure.