27 April 2026: Microsoft and OpenAI Reshape Their Alliance as China Blocks Meta’s Manus Deal

China has ordered Meta to unwind its multibillion-dollar Manus acquisition, Microsoft and OpenAI have restructured the partnership that underpinned the AI boom, and a leaked analyst note suggests OpenAI may be building a smartphone where AI agents replace the apps you use today.

China’s antitrust regulator has blocked Meta’s $2 billion acquisition of Manus, the autonomous AI agent platform, following a months-long investigation. Beijing ordered the deal to be unwound, dealing a significant setback to Mark Zuckerberg’s ambitions in the AI agents space. Manus launched in early 2025 and quickly became one of the most discussed autonomous AI tools, attracting Meta’s attention as the company sought to accelerate its own agent capabilities beyond what internal teams had built.

The decision reflects growing friction between Chinese authorities and Western technology companies seeking to acquire AI assets with potential dual-use or strategic significance. It also raises immediate questions about how Meta will now build out its agent strategy, having lost what would have been one of its most significant acquisitions in the field. Rivals including Google and Microsoft will be watching closely to see whether Meta pivots to internal development or returns with a revised approach.

OpenAI and Microsoft have restructured the partnership that helped launch the modern AI industry, replacing the original 2019 agreement with terms designed to give OpenAI greater independence. A joint statement confirmed the amended deal simplifies the relationship, adds long-term clarity on roles, and adjusts how the two companies share commercial upside. Under the new terms, Microsoft will no longer share revenue with OpenAI, though it retains exclusive rights to deploy OpenAI models commercially through Azure.

The original arrangement had been repeatedly expanded, ultimately providing OpenAI with billions in computing credits and early capital in exchange for Microsoft’s preferred commercial position. The restructure is widely interpreted as a necessary step before OpenAI’s anticipated public offering, removing the complexity of a revenue-sharing model that was designed for a smaller, earlier-stage organisation. For businesses currently using Azure OpenAI services, the day-to-day products and APIs are not expected to change in the near term.

OpenAI may be developing a smartphone in which AI agents perform the tasks that individual apps currently handle, with mass production potentially starting in 2028. An analyst report cited by TechCrunch describes a device that would represent a fundamental rethink of the smartphone interface. Rather than opening separate applications, a user would instruct an AI agent that handles the task autonomously, whether that involves booking travel, drafting messages, or researching a purchase decision. No official confirmation has come from OpenAI.

The concept would put OpenAI in direct hardware competition with Apple and Google, a remarkable shift for an organisation that began as a non-profit research lab in 2015. It also signals the direction the broader industry is heading. AI agents are moving from browser extensions and desktop tools toward becoming the primary layer through which people interact with technology. Consumers and small businesses watching this space should expect the devices and interfaces available in three years to look quite different from today.

Google and Kaggle have reopened registration for their free five-day AI Agents Intensive Course, running in June 2026, and anyone can sign up now via the Google AI blog. The programme, described as a “Vibe Coding” course, is designed to teach developers and technically curious non-coders how to build, test, and deploy AI agents from scratch. Previous cohorts filled quickly. The June edition will include updated materials reflecting the latest developments in multi-agent frameworks and tool use.

This is one of the most accessible entry points into AI agent development currently available at no cost. If you have been curious about building your own automation workflows or want to understand how modern AI agents actually function, the course runs across five days and is self-paced within the cohort window. Registration is open now.

A cybersecurity breach at Mercor, a company that supplies voice data for AI model training, has exposed approximately four terabytes of recordings from around 40,000 contractors. The stolen dataset includes voice samples used to train speech and language models, along with metadata that could be used to identify individual workers. Mercor has not yet confirmed the full scope of the incident or whether the data has been misused or offered for sale.

The breach is a reminder that the human infrastructure behind AI training carries significant personal data risks that receive relatively little public attention. Workers who provided voice recordings to AI companies may be unaware their data has been compromised. The incident is likely to draw scrutiny from UK and EU regulators, where rules on biometric data are considerably stricter than in the United States. Companies building with third-party training data providers should review their supply chain data handling policies.

Here is everything else worth knowing from today’s AI news.

• OSS Dirac agent tops TerminalBench — An open-source agent built on Gemini 3 Flash Preview scored 65.2% on TerminalBench, beating Google’s official score of 47.8% and the leading closed-source competitor. Source
• Meta signs a space-based solar deal — Meta has contracted with Overview Energy to supply power beamed from an orbital solar array, a first commercial step toward 24-hour renewable energy for AI data centres. Source
• Running local LLMs offline on a long flight — A developer guide explains how to run capable language models entirely on a laptop with no internet connection. Practical for privacy-conscious users or anyone working in low-connectivity environments. Source
• Chrome’s Prompt API gets fresh attention — Google’s built-in browser AI feature, which lets websites run small language models locally in Chrome without any server calls, is attracting new interest from developers building offline-first tools. Source
• AI should elevate thinking, not replace it — A widely read essay argues that heavy reliance on AI for reasoning and writing is narrowing rather than expanding human thought, and calls for users to treat AI as a thinking partner rather than a substitute. Source

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s afternoon update on developments in artificial intelligence, published every weekday afternoon.