AI week April 21-28, 2026: Google bets $40 billion on Anthropic, Copilot goes agentic, and cybersecurity gets a new front
Four developments left their mark on the AI market this week: Google's record investment in Anthropic, Microsoft Copilot Agent Mode becoming generally available in Office, a major AI cybersecurity initiative, and a deeper AWS-Anthropic partnership. Here's a quick, practical read on all of it.
1. Google bets up to $40 billion on Anthropic
On April 24, Google announced it will invest $10 billion in cash in Anthropic right away at a $350 billion valuation, with up to another $30 billion on the table if Anthropic hits performance targets. On top of that, Google Cloud will supply 5 gigawatts of compute capacity over the next five years. It is one of the biggest AI deals ever and firmly places Anthropic at the top of the frontier model field.
What does this mean for organizations? Access to frontier AI is getting deeper, but it's also concentrating further among a handful of players. For vendor strategy and risk management, it's important to design your AI stack so you're not locked into one model or one cloud.
2. Microsoft Copilot Agent Mode is GA in Word, Excel, and PowerPoint
On April 22, Microsoft announced that Copilot Agent Mode is generally available in Word, Excel, and PowerPoint on Windows, Mac, and the web. Copilot can now carry out multi-step actions in documents, sheets, and decks: from first draft to final version, with the user staying in control. Microsoft reports 67% growth in Excel engagement and 65% higher user satisfaction during the preview.
CribConnects AI take: This moves Copilot from "smart assistant" to "colleague who gets things done" in everyday Office work. Start redesigning your standard processes around reporting, financial analysis, and presentations now. Otherwise the value gets lost in scattered one-off prompts.
3. Anthropic launches Project Glasswing for defensive cybersecurity
On April 27, Anthropic announced Project Glasswing, a coalition that includes AWS, Apple, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, NVIDIA, Palo Alto Networks, and the Linux Foundation. Participants get early access to Claude Mythos Preview for vulnerability detection, black-box testing, and pentesting. Anthropic is adding up to $100 million in usage credits and $4 million in donations; the first public findings are expected within 90 days.
What to do? Security teams need to prepare for a world where both attackers and defenders use frontier models. Threat models, detection engineering, and red teaming all deserve a refresh, ideally this quarter.
4. AWS and Anthropic go deeper into the chip
The AWS Weekly Roundup of April 27 confirmed that Anthropic now trains its most advanced foundation models on AWS Trainium and Graviton, with co-engineering at the silicon level through Annapurna Labs. Claude Cowork is available in Amazon Bedrock, and a Claude Platform on AWS is in the works. Combine this with the Google deal and the picture is clear: Anthropic is positioning itself as a multi-cloud frontier player.
Conclusion of the week
Infrastructure, distribution, and the application layer all moved at once this week. For Dutch organizations, the three main questions are: how do you secure vendor diversity, how do you get real value out of agentic Office, and how do you professionalize your AI security? CribConnects AI helps you turn those questions into a concrete roadmap this week.
Want to know more? Book an AI Scan via cribconnects.ai and turn this week's insights into action.