AI News: July 17, 2026

1. Moonshot Releases Open Kimi K3, Narrowing the Gap With US Frontier Models

Moonshot AI. The Chinese lab released Kimi K3, an open mixture-of-experts model with roughly 2.8 trillion total parameters that activates 16 of 896 experts per token and posts benchmark scores approaching GPT-5.6 Sol and Fable 5. Coverage framed the release as a sign that the era of drastically underpriced Chinese models may be ending as Moonshot moves toward parity with leading US systems. The weights are available for developers to download and self-host. Source

2. Germany Puts AI Search Engines Under Media Law in First-of-Its-Kind Ruling

ZAK. Germany’s Commission for Licensing and Supervision classified AI search engines and chatbots as content providers for the first time, issuing enforceable rulings against Google’s AI Overviews and Perplexity under the State Media Treaty. Regulators held that AI-generated answers are independent content created by the providers themselves, so the Digital Services Act liability exemption does not apply, and accused Google of burying journalistic sources beneath its AI summaries. Both companies have one month to appeal the immediately enforceable orders. Source

3. Sakana AI Adds Nvidia Nemotron to Its Model-Orchestration Approach

Sakana AI. The Japanese lab said it is integrating Nvidia’s open Nemotron models into Fugu, its orchestrator that routes subtasks across a pool of language models and synthesizes their outputs through a unified API. Sakana positions the approach as evidence that “collective intelligence” from many coordinated models can rival single frontier systems while reducing vendor lock-in, though independent testers have questioned its speed and cost. The company reported its Fugu Ultra configuration performing on par with frontier models in its own benchmarks. Source

4. Elorian Raises $55M at a $300M Valuation to Build Visual AI Models

Elorian. The startup, founded by former Google DeepMind researcher Andrew Dai, raised a $55 million round at a $300 million valuation before shipping a product, with Menlo Ventures and Nvidia among its backers. Elorian is building visual AI models aimed at what Dai calls “visual AGI,” an area he argues has seen uneven progress. Dai said he prioritized strategic investors over higher valuation offers. Source

5. Oak Raises $60M Seed for an AI-Native Identity Platform

Oak. The identity startup raised a $60 million seed round led by Accel, Greylock, and CRV to build what it calls an AI-native identity operating system for governing access and permissions across human, machine, and AI agent identities. The pitch targets enterprises trying to control what autonomous agents can do inside their systems. The round reflects growing investor focus on securing agent access as companies deploy more automation. Source

6. Spectro Cloud Raises More Than $100M for AI Infrastructure Management

Spectro Cloud. The infrastructure company raised more than $100 million in a Series D from Growth Equity at Goldman Sachs Alternatives, AMD Ventures, Ericsson, and others to expand its platform for managing AI and Kubernetes infrastructure across enterprise and edge environments. The funding underscores demand for tooling that helps organizations operate AI systems at scale. It adds to a busy week of large enterprise AI infrastructure rounds. Source

7. Vendelux Raises $50M Series B for AI Event Intelligence

Vendelux. The B2B event-intelligence startup raised a $50 million Series B, bringing total funding to about $71 million, to expand its AI platform that helps companies decide which conferences and events to attend. The round reflects continued investment in applied AI for sales and marketing operations. Source

8. DoorDash Opens Its Ordering Platform to AI Agents With dd-cli

DoorDash. The company introduced dd-cli, a command-line tool in limited beta that lets developers and AI agents search stores, build carts, and place orders directly from the terminal. Co-founder Andy Fang said the tool exposes DoorDash’s ordering platform to agents as part of a broader “agentic commerce” push that also includes ChatGPT and Claude integrations. The beta is limited to US and Canadian macOS developers through a waitlist. Source

9. Roblox Adds AI Game Creation Inside Its Mobile App

Roblox. The gaming platform launched an AI-powered game-creation feature inside its mobile app, letting users generate and edit games from their phones rather than requiring the desktop Roblox Studio. The move extends generative creation tools to a much larger mobile audience. It reflects how consumer platforms are lowering the barrier to building interactive content with AI. Source

10. Cito Launches an Agent-Oriented Search Tool Over 236M Papers

Cito. A hybrid academic search tool that indexes more than 236 million papers and is built for AI agents launched on Product Hunt, aiming to give agents structured access to scientific literature. The product is part of a wave of tools designed to be consumed by autonomous agents rather than only human users. Source

11. Mayo Clinic Now Runs More Than 150 AI Models Across Its Operations

Mayo Clinic. The hospital system said it now runs more than 150 AI models and is testing a tool that generates patient summaries, orders records chronologically, and makes them easier for clinicians to search. The deployment is one of the larger real-world tests of AI inside a major healthcare provider. It illustrates how regulated health systems are moving from pilots to broad clinical use. Source

12. AMI Labs Founder Rejects ‘AGI’ and ‘Superintelligence’ Framing

AMI Labs. Founder Alexandre LeBrun said he avoids labels like “AGI” and “superintelligence” for his company’s technology, arguing the terms overstate what current systems do and distract from building useful products. The comments, made in a TechCrunch interview, add a skeptical founder voice to industry debates over how AI capabilities are marketed. Source