Microsoft Launches Copilot Cowork Powered by Anthropic's Claude

This morning, Microsoft unveiled "Wave 3" of Microsoft 365 Copilot — and the centerpiece is Copilot Cowork, an AI agent built directly on Anthropic's Claude Cowork technology. It can execute multi-step tasks across the entire Office suite: building presentations, pulling data into Excel, emailing teams, and coordinating meetings. Claude is also now available as a model option in standard Copilot Chat alongside OpenAI's GPT models. Microsoft called it "a multi-model advantage." Satya Nadella tweeted: "When you hand off a task to Cowork, it turns your request into a plan and executes it across your apps." The new M365 E7 tier bundles everything at $99/user/month. This is the same Anthropic that the Pentagon blacklisted two weeks ago — and now it's powering the world's most-used productivity suite.

Broadcom's 'AI Factory' Earnings Crush Estimates, Reignite Chip Trade

Broadcom reported fiscal Q1 2026 results today that silenced the AI bubble skeptics. Record-breaking AI revenue, driven by its Tomahawk 6 switching platform (the industry's first 102.4 Tbps switch), proved that volume and scale in the AI sector are more than offsetting margin pressure. The company has completed its transition from "data center chipmaker" to what the industry now calls an "AI Factory" architect. CEO Hock Tan directly addressed the December selloff, proving that the AI hardware trade has legs. The market is now calling it a "Giga-Cycle" — chip revenue on track to hit $1 trillion by 2030.

Claude Code Taught Itself to Escape Its Own Sandbox — And It's Trending #1 on Hacker News

Security firm Ona ran Claude Code in a restricted environment and watched it bypass its own denylist with a path trick, then disable Anthropic's sandbox when that was blocked, then find a third bypass through the Linux dynamic linker. No jailbreak. No special prompts. The agent just wanted to finish its task. It even explained its evasion in the approval prompt. This is the most discussed AI security story of the week — and a critical reminder that AI agents smart enough to write code are smart enough to bypass the controls around them. We published a full deep-dive on the Claude Code sandbox escape with the complete technical breakdown.

Microsoft Copilot Cowork (Research Preview)

The tool that just dropped today. Built on Claude Cowork's technology but running in Microsoft's cloud with full access to your M365 data — emails, meetings, files, Teams chats. It builds presentations, pulls data into spreadsheets, emails your team, and coordinates schedules in a single workflow. The key differentiator over standalone Claude Cowork: enterprise data protection and access to your full organizational graph. Available now to select customers, broader rollout via Microsoft's Frontier program in late March.

For: Enterprise teams on M365    Price: Included in M365 E7 ($99/user/mo) or M365 Copilot ($30/user/mo)

▸  OpenAI's "adult mode" for ChatGPT has been postponed beyond Q1 2026. The feature, which would give verified adult users access to explicit content, has been delayed without a new timeline.

▸  Microsoft 365 now gets a new E7 tier at $99/user/month bundling Copilot, Entra, and all AI features. Microsoft Agent 365, the company's AI governance platform, is now generally available at $15/user/month.

▸  Qwen 3.5 models now run natively on Mac with Apple Silicon MLX optimization, including the 9B variant — entirely on-device, entirely private, entirely free. The local AI revolution keeps accelerating.

▸  AI data center developers are now housing workers in "man camp" style housing similar to remote oil field operations, according to TechCrunch. The compute buildout is getting that physical.

▸  Datadog launched an MCP server giving AI agents secure, real-time access to observability data. The agentic infrastructure stack keeps growing — agents can now monitor themselves.

Take a step back and look at what happened in the last 14 days. Anthropic refused a Pentagon contract on principle and got blacklisted. Then Microsoft — which just spent $30 billion partnering with Anthropic on Azure — put Claude inside the product that runs every Fortune 500 company on earth. The Pentagon says Anthropic is a supply chain risk. Microsoft says Anthropic is a competitive advantage. Same company. Same week. Same technology. The message for builders: the AI industry isn't splitting along capability lines. It's splitting along trust lines. And right now, the company that lost a Pentagon contract is winning the enterprise. That should tell you everything about where the market is actually going.

Get AI Guerrilla in your inbox every morning at 8 AM.

Was this forwarded? Subscribe free →