OpenAI, Anthropic and Google team up on shared agent standards
OpenAI, Anthropic, Google, Microsoft, Amazon and others are joining forces on a new industry group called the Agentic Artificial Intelligence Foundation. The aim is to build shared standards so AI agents can reliably connect to the enterprise apps they work inside.
The Linux Foundation will host the project, which starts with three open tools already used across the industry: Anthropic’s Model Context Protocol, OpenAI’s Agents.md format, and Block’s Goose, a lightweight local agent. MCP is gaining the fastest traction because it gives models a consistent way to talk to APIs across apps like Slack, Google Workspace and internal systems.
But CIOs say security is a growing concern, especially around prompt injection and patching vulnerabilities across different tools. The new group will set rules for how companies contribute, fix issues and align on safe agent behaviour. Over time, shared standards could help agents shift from experimental tools to dependable workplace infrastructure.
Supported by AssemblyAI
Most teams juggle multiple APIs for transcription, diarization, redaction, and topic detection. It slows development and adds cost.
AssemblyAI combines it all in one API:
Multilingual Speech-to-Text, async or streaming
Speaker detection, topic detection, and PII redaction
Simple, usage-based pricing: $0.15/hr, no lock-in
No complex setup. No long onboarding. Just a reliable stack you can test on real audio workloads.
Build voice features, internal tools, or AI products—ship faster without vendor headaches or broken integrations.
DeepSeek is training its next model on smuggled Nvidia chips
DeepSeek has been developing its next model on thousands of Nvidia Blackwell chips that the U.S. bans from export to China. The chips were routed through overseas data centers, lifted from servers, and shipped into China in pieces, according to people familiar with the process. It shows how difficult U.S. chip controls are to enforce in a sprawling global supply chain.
The access gives DeepSeek an edge. Blackwell is far ahead of domestic alternatives, especially for the sparse attention methods the company is betting on. The approach lowers inference costs but slows development, and the next model is still being prepared.
Smuggling has become common as Nvidia hardware grows more valuable. Regulators are trying to tighten enforcement, but the distributed reseller network makes tracking difficult. For now, DeepSeek still relies heavily on Nvidia even as Beijing pushes for homegrown chips.
Americans are using and worrying about AI more than ever
A new data from AI governance nonprofit Fathom shows nearly seven in 10 Americans have used AI, but concerns about jobs, education and safety are rising.
Trust in federal regulators is low. Only 42 percent trust government agencies to create proper guardrails. Independent experts rank highest at 55 percent, followed by tech companies at 50 percent.
Usage keeps climbing. Eighty two percent of respondents say they have seen or heard of AI, and almost half of users now rely on it daily or weekly. But anxiety is growing too. Thirty three percent say they are worried about AI, up from 30 percent last year, and nearly half expect significant job losses.
More than 75 percent of people want clear rules to protect children using AI. Most still interact with AI through search engines and social platforms rather than standalone tools. The findings will sit at the center of new governance proposals expected next year as states consider how to regulate AI.
Wall Street hedges the AI trade for 2026
Markets are still bullish on AI, but the tone is shifting. After a year when hedging barely mattered, investors are adding protection as macro uncertainty and AI optimism collide. JPMorgan strategist David Kelly says 2026 is about reducing risk rather than chasing returns.
Most firms expect moderate gains. Bloomberg’s survey puts the S&P 500 at 7,269 by year end, roughly 6 percent above today. It is tame compared to the rally since 2023, but confidence in AI earnings remains high. Bank of America says tech stocks are not in bubble territory but warns of an “AI air pocket” ahead.
Investors feel pressure to stay fully invested. One RBC client called himself a “fully invested bear,” holding AI positions because the career risk of missing further upside is too high.
The key variable is the labor market. Weakness could hit retail investors, who make up about a quarter of daily trading. As Kelly put it, the biggest risk in 2026 might be the one no one sees coming.