Presented by
Good morning, AI enthusiasts. China just made it clear that AI acquisitions are geopolitical moves, not just business deals. Two founders of Manus AI can't leave the country after Meta agreed to buy the startup for $2B. Beijing's message is hard to miss.
Any AI company touching Chinese talent or IP may find its future negotiated not in boardrooms, but with regulators who can restrict your movement. For Western companies betting on global AI talent, that's a new kind of risk.
In today's recap:
Meta's $2B deal leaves Manus founders stuck in China
Figure robot makes its White House debut
Generate background music with Lyria 3 Pro
ARC-AGI-3 humbles frontier AI at under 1%
4 new AI tools, prompts, and more
META
China bars Manus founders from leaving country
Bloomberg
Recaply: China just barred the two co-founders of Manus AI from leaving the country, as Beijing reviews Meta's $2B purchase of the Singapore-based AI startup.
Key details:
Manus CEO Xiao Hong and chief scientist Ji Yichao were called to a meeting in Beijing with China's top economic planning agency. After the meeting, they were told they couldn't leave the country, though they remain free to travel within China.
Meta agreed to buy Manus in December 2025, with the startup valued at $2B to $3B at the time. China's commerce ministry said earlier this year it would investigate the deal.
The Financial Times first reported the story, noting that Manus is now looking for law firms to help resolve the situation.
No deadline has been set for the review. The founders remain in China with no confirmed end date.
Why it matters: The Manus deal looked clean when it was announced in December. Meta was buying a Singapore-based AI startup, and China wasn't supposed to be part of the story. But Beijing called the founders in and told them they couldn't leave. That's not a typical business dispute. It's a warning: any AI company with Chinese ties could find its founders stuck the moment a regulator decides to take a closer look.
PRESENTED BY MORNING BREW
Wake up to better business news
Some business news reads like a lullaby.
Morning Brew is the opposite.
A free daily newsletter that breaks down what’s happening in business and culture — clearly, quickly, and with enough personality to keep things interesting.
Each morning brings a sharp, easy-to-read rundown of what matters, why it matters, and what it means to you. Plus, there’s daily brain games everyone’s playing.
Business news, minus the snooze. Read by over 4 million people every morning.
FIGURE
Figure humanoid debuts at White House with First Lady
APC
Recaply: Figure's humanoid just debuted at the White House with First Lady Melania Trump, walking alongside her into the East Room and greeting guests in 10 languages during an education summit.
Key details:
The robot, named Figure 03, joined Melania Trump at the Fostering the Future Together summit. The event brought first spouses and education leaders from multiple countries to the White House over two days.
This is the first time a humanoid robot has escorted a First Lady at an official White House event.
Figure 03 introduced itself to the crowd: I'm Figure 03, a humanoid built in the United States of America. It then greeted guests in 10 languages. The AP, Fortune, and CNET all covered the event.
The summit concluded on March 25, 2026. Figure 03 appeared at the final day's roundtable.
Why it matters: Figure didn't just build a robot for a warehouse. It walked one into the White House. By placing Figure 03 next to the First Lady at a global education summit, the company is going after federal contracts and government credibility. That's harder than pitching a deck, but other robotics companies can't easily copy it. Whether this converts into real contracts is still the open question.
TUTORIAL
Generate original background music for videos with Lyria 3 Pro
Recaply: In this tutorial, you will learn how to use Lyria 3 Pro to create original background music for videos, podcasts, or social content, without licensing costs or royalty-free loops.
Go to Google AI Studio at aistudio.google.com or open the Gemini app on your phone. Sign in with a Google account and navigate to the music generation tool powered by Lyria 3 Pro.
Describe the track you want by specifying mood, tempo, genre, and structure. For a YouTube video, try: Upbeat background music with an intro, two verses, and a clean outro, no vocals, 2 minutes long.
Generate the track and listen back. If it doesn't match, adjust specific elements in your prompt. Swap upbeat for calm and focused, change the tempo, or add minimal percussion to refine the sound.
For longer tracks, use structural prompting and generate each section separately. Prompt for an intro, a main loop, and an outro, then combine them in your video editor.
Export the audio and bring it into your editor or podcast software. All Lyria 3 Pro outputs include Google's SynthID watermark, which is imperceptible to listeners but identifies AI-generated content.
Pro tip: Use ProducerAI at producer.ai for a more collaborative music creation flow built on Lyria 3 Pro. It's free and works well for longer projects like full episodes or YouTube series.
ARC PRIZE
ARC-AGI-3 launches, frontier models score under 1%
Arc Prize
Recaply: ARC Prize just launched ARC-AGI-3, a new benchmark testing agentic AI on interactive reasoning environments, with all frontier models scoring under 1% while humans score 100% without training or instructions.
Key details:
ARC-AGI-3 uses RHAE scoring, measuring how efficiently an AI agent solves novel interactive environments compared to the second-best human baseline, with scores calculated as efficiency squared.
The best frontier models, including Gemini 3.1 Pro Preview at 0.37%, GPT-5.4 High at 0.26%, and Claude Opus 4.6 Max at 0.25%, all scored under 1%, with each task costing AI systems $2K to $9K in compute.
Benchmark creator Francois Chollet said that if a system is truly AGI, it should be able to make its own harness for a new task, or just internalize it, framing action efficiency as the clearest test of genuine reasoning.
ARC-AGI-3 launched March 26, 2026, with 25 environments now public and a Kaggle competition running on 55 private environments.
Why it matters: Every major AI benchmark of the last two years has been saturated within months. ARC-AGI-2 was cracked, so Chollet designed a harder one. The 0.2-0.3% scores show that what AI has mastered is pattern matching on training data, not reasoning from scratch on something genuinely new. If real AGI means a system that figures out a task it's never seen before as efficiently as a human, we aren't there yet.
TOOLS
Trending AI Tools
🎵 Lyria 3 Pro - Google's music generation model
🤖 Ghostwriter - Sierra's agents-as-a-service platform
⚙️ Linear Agent - Linear's AI agent for issue triage and project management
💻 Claude - Anthropic's work tool integrations now available on mobile
NEWS
What Matters in AI Right Now?
Tufts University released the first American AI Jobs Risk Index, estimating 9.3 million US jobs could be displaced in the next 2-5 years, with $757B in annual wages at the midpoint of its 2.7M to 19.5M displacement range.
Bernie Sanders and AOC introduced a bill to pause construction of new datacenters, arguing the energy demands of AI infrastructure are unsustainable, with critics noting the move would shift compute buildout to countries with fewer environmental restrictions.
Anthropic published its Economic Index March 2026 report, tracking how Claude is used across the economy using privacy-preserving data analysis from February 2026, building on its January framework of economic primitives.
OpenAI launched a public Safety Bug Bounty program, expanding its existing security program to accept submissions for AI abuse and safety risks, with rewards for meaningful misuse risks that don't meet traditional security vulnerability thresholds.
Bland AI introduced Norm, a voice AI agent builder that creates production-ready voice agents from a single prompt, letting teams describe what they want and generate a deployable agent without manual configuration.
Perplexity CEO Aravind Srinivas argued that AI-driven job displacement isn't inherently harmful because most people dislike their jobs anyway, framing automation as an opportunity for entrepreneurship rather than a crisis.
The US Department of Labor launched "Make America AI-Ready," a free AI literacy initiative letting workers access a foundational AI course by texting "READY" to 20202, as part of the Trump administration's workforce AI strategy.
🧡 Enjoyed this issue?
🤝 Recommend our newsletter or leave a feedback.
How'd you like today's newsletter?
Cheers, Jason
