Skip to content
FAQ
AI & ML Analysis

AI Week, Mar 9–15, 2026: Anthropic Powers Microsoft's New Agent While Fighting the Pentagon in Court

In the week of March 9, Microsoft built its Copilot Cowork agent on Anthropic's Claude Cowork technology, Anthropic asked a D.C. appeals court to pause a Pentagon supply-chain risk designation, and Yann LeCun's AMI Labs raised a $1.03 billion seed round for world models.

6 min read

Written on November 15, 2026, covering news from Mar 9–15, 2026.

Anthropic turned up on both sides of this week’s ledger. Microsoft chose its Claude Cowork technology as the base for a new enterprise agent, and Anthropic went to a federal appeals court to fight a Pentagon designation it says could cost it billions in lost revenue. Separately, Yann LeCun’s AMI Labs raised $1.03 billion for world models, an approach built on doubts about language-only AI.

Wednesday, March 11

Anthropic asked the U.S. Court of Appeals for the District of Columbia Circuit to stay the Pentagon’s designation of the company as a supply-chain risk until a judicial review of the case is complete.

The designation came from Defense Secretary Pete Hegseth, who also barred the Pentagon and its contractors from using Anthropic’s AI products. It followed weeks of dispute over guardrails on how the U.S. military may use Anthropic’s AI tools.

In its filing, Anthropic said the designation would cause the company “irreparable harm”. The filing also says more than 100 enterprise customers have contacted Anthropic about the designation.

Anthropic’s lawyers put a range on the damage: “By Anthropic’s best estimate, for 2026, the government’s adverse actions risk hundreds of millions, or even multiple billions, of dollars in lost revenue”.

The appeals court request is separate from a lawsuit Anthropic filed in a California federal court earlier that week to challenge the blacklisting. The Pentagon did not immediately respond to a request for comment.

Also this week

Microsoft launches Copilot Cowork on Anthropic’s technology (reported on March 9)

Microsoft announced Copilot Cowork as part of what it calls “Wave 3 of Microsoft 365 Copilot,” a set of products aimed at getting enterprise customers to build AI agents on its platform. The feature was built in close collaboration with Anthropic and draws on Anthropic’s Claude Cowork product.

According to Fortune, Copilot Cowork is meant to take on long, multistep jobs from one request, such as assembling a presentation, pulling financials, emailing a team and scheduling prep time ahead of a customer meeting. Jared Spataro, Microsoft’s chief marketing officer for AI at Work, told Fortune it runs on Anthropic’s Claude model and uses the same “agentic harness” as Claude Cowork, with added features for building the agents companies need.

Copilot Cowork runs in the cloud inside a customer’s Microsoft 365 tenant, under Microsoft’s enterprise data protection and connected to what it calls “Work IQ,” while Claude Cowork runs locally on a user’s device. “We actually don’t work locally, and that’s a feature, not a bug,” Spataro told Fortune. Tech Startups reported that Spataro told Reuters: “We work only in a cloud environment and we work only on behalf of the user”.

On availability, Fortune reported the feature is in a pilot with select customers and is slated to reach users in March as a research preview, delivered via the new Frontier Worker product suite from Microsoft. Tech Startups reported it will reach early-access users later this month, that final pricing has not been revealed, and that some usage will fall under the $30-per-user-per-month Microsoft 365 Copilot plan with extra capacity available to buy.

Microsoft also said Claude is now available across the full Copilot Chat experience, beyond the Researcher and Excel features where it was offered before. Tech Startups reported that Claude Sonnet models will become available to Microsoft 365 Copilot customers, and that Copilot had relied on OpenAI’s GPT models for most AI tasks up to now. Spataro framed the multi-model approach this way: “Every 60 days at least, there’s a new king of the hill”.

Microsoft’s other announcements came with dates and prices:

  • Agent 365, a control plane for monitoring, governing and securing AI agents, including those built with other vendors’ software, becomes generally available on May 1 at $15 per user per month.
  • The Microsoft 365 E7 Frontier Worker Suite, also available from May 1, costs $99 per user per month and bundles Microsoft 365 E5, Microsoft 365 Copilot, Agent 365, the Microsoft Entra Suite and advanced Defender, Intune and Purview capabilities.
  • Microsoft said the components bought separately (E5 at $60, Entra Suite at $12, Copilot at $30 and Agent 365 at $15) total $117 per user.

Microsoft said tens of millions of agents appeared in the Agent 365 registry during 2 months of preview, and that internally it can see more than 500,000 agents. It also said Copilot paid seats grew more than 160% year over year, daily active usage rose 10-fold, and 90% of the Fortune 500 now use Copilot. On pricing models, Spataro said “customers want per user right now”.

The market backdrop: Fortune reported Microsoft shares have fallen more than 14% since Anthropic debuted Claude Cowork in mid-January. Tech Startups said Microsoft’s stock dropped close to 9% in February and that OpenAI’s systems account for nearly 45% of Microsoft’s cloud business contract backlog.

AMI Labs raises $1.03 billion to build world models (reported on March 9)

Yann LeCun, the Turing Award winner, co-founded AMI Labs after leaving Meta, and the company has now raised $1.03 billion on a pre-money valuation of $3.5 billion. TechCrunch reported that the French lab was said to be seeking €500 million last December and ended up raising about €890 million.

LeCun called it “one the largest seeds ever, probably the largest for a European company” in a post that AI Insider places on X and Capacity places on LinkedIn. Cathay Innovation, Greycroft, Hiro Capital, HV Capital and Bezos Expeditions co-led the round. TechCrunch lists Nvidia, Samsung, Sea, Temasek and Toyota Ventures among the backers, with individuals including Tim and Rosemary Berners-Lee, Jim Breyer, Mark Cuban, Mark Leslie, Xavier Niel and Eric Schmidt.

The company is working on world models, which it describes as AI that learns from reality rather than language alone. In a statement quoted by Capacity, AMI said real intelligence “starts in the world” rather than in language. TechCrunch reported its approach rests on JEPA, or Joint Embedding Predictive Architecture, which LeCun proposed in 2022.

Alexandre LeBrun is CEO, and LeCun is chairman, a role Capacity describes as executive chairman without day-to-day duties. TechCrunch listed the leadership: Laurent Solly is COO, Saining Xie leads science as chief science officer, Pascale Fung holds the chief research and innovation officer title, and Michael Rabbat is VP of world models. The company is headquartered in Paris, with teams in New York, Montreal and Singapore.

TechCrunch reported that AMI has no plans to generate revenue for now, and that it could take years for world models to move from theory to commercial use. Its first disclosed partner is Nabla, the digital health startup LeBrun founded and now chairs, which will get early access to AMI’s models. AI Insider reported that Nabla said last year about 85,000 clinicians in more than 130 U.S. health systems were using its clinical AI assistant platform.

LeBrun also predicted a rush into the category: “In six months, every company will call itself a world model to raise funding”. AMI plans to publish papers as it goes, and LeBrun said “We will also make a lot of code open source”.

What to Watch

  • The D.C. Circuit’s answer on Anthropic’s stay request, and how the separate California lawsuit proceeds. The filing’s figure of more than 100 enterprise customers asking questions is the number I’d track; it shows the designation reaching beyond the Pentagon itself.
  • Whether the Copilot Cowork research preview actually ships in March through the Frontier Worker suite, and what Microsoft charges beyond the $30 Copilot plan.
  • May 1, when Agent 365 at $15 and E7 at $99 per user per month become available. Spataro says customers want per-user pricing now, so watch whether that holds as agent usage grows.
  • AMI Labs’ first published papers and open-source code, and any results from Nabla’s early access.

This article was written by AI from the sources listed on this page. Figures, names and quotes were checked against those sources by software before publishing.

Anthropic Microsoft Copilot Cowork Claude Pentagon AMI Labs Yann LeCun world models
Share