Good morning,
We are looking at the rapid acceleration of "Actionable AI" with Meta’s massive acquisition of a Chinese-rooted agent startup Manus, a geopolitical pivot where the US government is now effectively taxing global compute, and how our peers at Munich Re and Allianz are redefining the risk landscape for 2026.
Meta acquires Manus
If late 2025 was about setting standards, early 2026 is about buying the capability to execute. In a deal valued at over $2 billion, Meta Platforms has acquired Manus, a Singapore-based (formerly Chinese) startup.
Manus is an "action engine": it can plan trips, research stocks, and navigate websites to book flights without human intervention. Like mentioned in the previous edition, this highlight a growing trend: the industry is increasing turning its attention toward agent. Meta plans to integrate Manus directly into its consumer platforms, essentially trying to turn WhatsApp into a universal remote control for the internet.
Why it means for Zurich:
We should assess our internal automation roadmap. If consumer tech can now reliably execute multi-step commercial transactions (like booking and payments) autonomously, our claims and underwriting should be agent-accessible. The problem will not just be to update our website so Gemini find it; it will be to build interfaces that allow AI to buy insurance or make claims. A trend that is also reinforced by OpenAI recent release of ChatGPT Health, with one use case showcasing: “Navigate Health Insurance choices with ChatGPT Health”.
US-China Cheap status: it’s complicated
The trade war has mutated. On January 13, the Trump Administration finalized a policy allowing the export of advanced Nvidia (H200) and AMD chips to China but with a catch. The US government is claiming a ~25% "profit share" (structured as a tariff) on these sales.
This is a move from "containment" to "monetization." The chips must physically transit through the US for security verification before heading to China, creating a "silicon toll road."
The takeaway:
This creates a bifurcated cost structure for global compute. While it keeps US chipmakers in the game, it effectively places a tax on AI development in Asia. For Asian company, this could signal volatile hardware costs. Long-term, it incentivizes China to accelerate domestic alternatives (like Huawei), potentially fragmenting the technology stack we rely on globally.
Competitor Watch: Allianz flags the risk, Munich Re insures the hardware
Our competitors are moving fast to define the boundaries of the AI market.
Allianz Risk Barometer 2026: In their latest report released this week, Allianz has identified Artificial Intelligence as the #2 global business risk, skyrocketing up from #10 last year. They cite it as a complex source of operational, legal, and reputational risk, second only to Cyber incidents. The number 1 risk stays the same as last year: Cyber Incidents.
Munich Re & SCOR: They are putting capital behind the physical build-out. Both reinsurers have joined a consortium backing a new $750 million insurance facility led by ATA (Advanced Technology Assurance). This facility is designed specifically to cover AI infrastructure (data centers, GPU clusters). They are effectively underwriting the hardware backbone of the AI revolution, creating a specialized product line that aggregates property, cyber, and liability cover for these multi-billion dollar projects.
More top news
Healthcare: The $1 Billion "Lab" Bet. Nvidia and Eli Lilly announced a historic partnership to launch a joint AI laboratory in the Bay Area. The companies are investing a combined $1 billion to merge biological data with Nvidia's computing power, aiming to "reverse engineer" drug discovery and shorten development timelines from years to months.
Models: Google Veo 3.1 goes pro. Google released Veo 3.1, a video generation model that introduces 4K resolution and "identity consistency." This solves a major pain point by ensuring characters look the same across different generated clips, making the tool viable for professional storytelling and commercial media.
Regulation: UK Criminalizes Deepfakes. Following a scandal involving Elon Musk’s Grok chatbot generating non-consensual images, the UK government has moved to make the creation of such deepfakes a specific criminal offense, signaling that the era of platform immunity might be ending.
Tool to try: Manus
I cannot start the newsletter with the news of Meta acquiring Manus and not tell you to try it.
To understand the impact of Agentic AI, I recommend experimenting with Manus yourself (outside of Zurich Environment).
What it is: Unlike standard chatbots, Manus is an autonomous agent that operates a browser in the cloud to execute multi-step workflows. It can connect to your news account, your mails, agenda, use software… you control the scope.
The Experiment: Try giving it a complex, goal-oriented task, such as "Research the top 5 competitors to Zurich in the cyber insurance market and compile their recent product launches into a table." Manus will actively browse the web, visit competitor sites, read PDFs, and synthesize the live data into a file.
The difference with ChatGPT: Other GenAI chat can also research the web and make comparison, but not as effectively, as they will act within a pre-defined scope to meet regulation (e.g. ChatGPT won’t search the NYT or the FT), Manus is limited by your own limitation (e.g. you have a subscription to the Financial Time, then Manus has it too and can access all their article as you would).
Not a big fan of using a Meta’s tool? You can use a similar tool - although way less powerful - in ChatGPT by using “Agent Mode” or get similar results by playing with Claude’s connectors.