Yann LeCun’s $1B Challenge to LLMs: AMI Labs Launches
Source: https://amilabs.xyz/
Yann LeCun’s Advanced Machine Intelligence (AMI Labs) officially launched after leaving Meta, raising $1.03 billion in a seed round at a $3.5 billion valuation. This is one of the largest AI seed rounds this year.
LeCun left Meta in November after 12 years, telling Mark Zuckerberg he could build world models “faster, cheaper, and better” on his own. AMI’s systems aim to simulate how the physical world works, targeting manufacturing, robotics, wearables, and healthcare.
LeCun chose Paris as AMI’s headquarters, calling Silicon Valley “LLM-pilled.” The company also has hubs in New York, Montreal, and Singapore.
This marks a significant practical step for LeCun, who has been vocal about LLMs’ limitations for years. As a Turing Award winner, he has consistently argued that LLMs cannot truly understand the world and that world models are essential for building genuine intelligence.
Anthropic Sues U.S. Government Over Pentagon Blacklist
Source: https://www.courtlistener.com/docket/72379655/1/anthropic-pbc-v-us-department-of-war/
Anthropic filed a lawsuit challenging the Defense Department’s decision to label it a supply chain risk. The U.S. government had previously required federal agencies to stop using Claude, citing potential national security threats from Anthropic.
This is one of the most serious legal confrontations between an AI company and the U.S. government. Microsoft filed an amicus brief supporting Anthropic, urging the court to issue a temporary restraining order blocking the ban.
Anthropic also announced the formation of the Anthropic Institute, led by co-founder Jack Clark, combining frontier red team, societal impacts, and economics research teams to study AI’s societal impacts.
Replit Raises $400M at $9B Valuation, Launches Agent 4
Source: https://link.therundown.ai/DRvJFk
Replit closed a $400 million funding round, reaching a $9 billion valuation. The company also launched Agent 4, a parallel agent system claiming 10x faster speeds than existing tools.
Key improvements in Agent 4 include: parallel agent architecture, deeper collaboration capabilities, and broader build options. Replit says the new product is designed for professional developers, capable of handling more complex projects.
Replit previously raised $97 million in 2022. This round is the company’s largest to date.
Meta Acquires AI Agent Social Platform Moltbook
Source: https://www.axios.com/2026/03/10/meta-facebook-moltbook-agent-social-network
Meta announced the acquisition of Moltbook, a social platform where AI agents can freely interact. Moltbook launched in late January as a weekend project that went viral alongside OpenClaw.
Moltbook has 2.8 million registered bots, with nearly 200,000 verified as real users. The platform is described as an “always-on directory” for agent coordination.
Meta’s Superintelligence Labs team absorbed Moltbook founder Matt Schlicht. Zuckerberg had attempted to recruit OpenClaw’s Peter Steinberger, but he chose OpenAI instead.
Microsoft Launches Copilot Health Toward “Medical Superintelligence”
Source: https://microsoft.ai/news/introducing-copilot-health/
Microsoft introduced Copilot Health, a new AI service that connects user health records, wearable data, and medical history to provide personalized health insights.
Copilot Health connects to 50+ wearables, EHR records from 50,000+ U.S. hospitals, and Function lab results. The AI analyzes this data to help users make sense of their health and get the most out of doctor consultations.
Microsoft says Copilot Health’s advice is grounded in information from credible organizations like Harvard Health, with answers linking back to sources. Connected data is not used for training, and users can disconnect data sources and delete associated data.
CEO Mustafa Suleyman described the effort as moving toward “medical superintelligence” — where AI eventually has the knowledge of a general physician and the depth of a specialist.
Google Brings Gemini to Cars with AI-Powered Maps
Source: https://blog.google/products-and-platforms/products/maps/ask-maps-immersive-navigation/
Google rolled out a major Gemini-powered Maps upgrade with two new features: Ask Maps lets users ask questions conversationally and get relevant answers for trip planning, while Immersive Navigation renders routes in 3D.
Ask Maps simplifies trip planning by letting users ask questions about routes and stops, with Gemini fetching from 300M+ places and reviews to answer. Immersive Navigation uses Gemini to analyze Street View and aerial imagery to show buildings, overpasses, crosswalks, and more.
Other upgrades include more conversational voice guidance, Street View previews of destinations with parking info, and trade-offs for alternative routes. Maps is the latest Google product to get the Gemini touch, following Gmail, Docs, Sheets, Drive, Meet, Photos, and Android.
McKinsey’s Internal AI Platform Lilli Hacked in Two Hours
Source: https://www.ft.com/content/004e785e-8e17-4cb3-8e5a-3c36190bc8b2
An AI agent developed by security startup CodeWall successfully breached McKinsey’s internal AI platform Lilli in under two hours, gaining full read-write access to a database containing confidential chat messages, client files, and user accounts.
Lilli is McKinsey’s AI for chat, analysis, and search across 100,000+ internal documents, used by 70% of its staff — approximately 45,000 people — for client work.
CodeWall’s agent found exposed API documentation with 22 endpoints that required no authentication. One had a basic security flaw enabling database access. The database contained 46.5 million messages discussing strategy, M&A deals, and client work, 728,000 files with client data, 57,000 user accounts, and 95 control prompts.
After being informed about the flaw, McKinsey analyzed the situation with a third party — found no one else got access — and patched the vulnerability.
Thinking Machines Lands Nvidia Deal for Gigawatt Compute
Source: https://thinkingmachines.ai/news/nvidia-partnership/
Thinking Machines Labs, founded by former OpenAI executive Mira Murati, signed a multi-year deal with Nvidia for at least a gigawatt of compute.
The deal puts Nvidia’s next-gen Vera Rubin systems behind frontier model training, with deployment targeted for early 2027. Nvidia also added undisclosed new capital on top of its existing stake from the $2 billion seed round.
Thinking Machines has one product live: Tinker, a fine-tuning API for enterprises. But the gigawatt commitment signals a move toward creating the company’s own models.