This post is part of my Weekly AI Signals series—a curated look at the moments that matter once the noise fades.

Introduction

As we step from 2025 into 2026, I want to pause on the final week of the year — not because it was loud, but because it was revealing.

Three developments stood out. China released draft rules aimed at emotionally engaging AI systems. SoftBank moved to strengthen its position in digital infrastructure. And Meta acquired a company focused on autonomous AI agents.

Individually, these stories are fascinating. Taken together, they suggest something deeper: AI is moving into a phase shaped by governance, physical scale, and questions of agency. Let’s walk through what happened — and what it might mean for where we’re headed.

1. China Moves to Regulate Human-Like AI Companions

Source:

China outlines rules to regulate human-like AI companion apps

On 27 December 2025, China’s Cyberspace Administration published draft rules aimed at AI services that simulate human behaviour and form emotional interactions. These drafts are open for public comment through late January 2026.

According to SiliconANGLE, the proposed rules focus on AI companions that feel persistent and human-like. They would require that such systems:

• Clearly disclose they are AI, using visible prompts rather than appearing human.
• Prompt users to take breaks after extended continuous use (the article notes a two-hour continuous threshold).
• Assess emotional states and take action if a user shows unhealthy dependence.
• Require human review if users express suicidal or self-harm ideation.
• Prevent harmful content, including violence, crime, or manipulative behaviours.

SiliconANGLE also connects this move to similar efforts in the U.S., such as California’s SB-243, which introduces usage reminders and other safety requirements for minors interacting with AI companion apps.

Why this matters

What’s notable here is how regulation is framing the issue: not just in terms of content accuracy or privacy, but in terms of user wellbeing and emotional interaction design. When AI systems behave in ways that feel relational, regulators are beginning to treat that as a distinct risk vector worth managing.

For creators, this is a reminder: design intent and user experience matter for governance just as much as data handling or output quality.

2. SoftBank and DigitalBridge: Infrastructure as Strategy

Source:

SoftBank Group to Acquire DigitalBridge for $4 Billion to Scale Next-Gen AI Infrastructure

On 29 December 2025, SoftBank Group announced it had signed a definitive agreement to acquire DigitalBridge Group for about $4 billion. DigitalBridge is a global investor and operator of digital infrastructure, including data centres, fibre networks, and connectivity systems — the physical backbone for large-scale cloud and AI workloads. [2]

What the accessible source shows

• SoftBank will acquire DigitalBridge Group, Inc. for roughly $4.0 billion in cash. [2]
• DigitalBridge focuses on data centres, cell towers, fibre networks and edge infrastructure. [2]
• SoftBank’s announcement explicitly connects the acquisition to supporting next‐generation AI infrastructure by expanding capacity and connectivity. [2]

Why this matters

While AI models and algorithms get most of the attention, large-scale AI depends on the world’s physical infrastructure: sites with power, cooling, fibre, and network capacity.

By bringing DigitalBridge into its fold, SoftBank is investing in the substrate that enables AI compute at scale — not just the software layer.

For builders and architects of AI systems, this highlights a reality: compute and connectivity at scale are strategic assets, not utilities.

3. Meta Acquires Manus: A Bet on Autonomous Agents

Sources:

Meta buys startup Manus in latest move to advance its artificial intelligence efforts

Meta to acquire AI startup Manus in deal valued at over $2 billion

On 30 December 2025, Meta announced it was acquiring Manus, a fast-growing AI startup focused on autonomous task-executing agents, in a deal valued at more than $2 billion (reported by Euronews and AP). Both outlets describe the acquisition, the technology focus, and changes to Manus’s ownership post-deal. [4]

What the accessible reporting confirms

• Meta is acquiring Manus, a Singapore-based AI startup. [4]
• The deal is valued at more than $2 billion, according to multiple sources, and is part of Meta’s push to expand AI across its products. [4]
• Manus developed a general-purpose AI agent that handles tasks like research, coding, and business workflows. [4]
• The platform had grown rapidly, surpassing $100 million in annual recurring revenue within eight months. [4]
• Meta confirmed that any previous Chinese ownership interests would be exited and Manus’s operations in China would wind down. [4]

Why this matters

Manus isn’t just another chatbot. Its technology represents a class of systems that execute multi-step tasks autonomously, not just react to prompts. That capability — turning high-level goals into sequences of action — is becoming central to what many companies call AI agents.

Meta is integrating this capability to complement its broader AI strategy, which includes both consumer-facing AI tools and enterprise offerings.

The Pattern That Emerges

Taken together, these developments suggest a shift:

• Emotional UX as a governance issue. China’s draft rules treat relational AI as a distinct regulatory focus.
• Infrastructure matters as much as models. SoftBank’s deal highlights that data centres and networks are strategic for AI’s future.
• Agents are the next frontier. Meta’s Manus acquisition suggests AI that acts — not just responds — is becoming a primary frontier.

These aren’t isolated headlines. They’re signals about where AI’s future is being shaped — by policy, physical capacity, and autonomous capabilities.

What I’ll Be Watching in 2026

• Whether other countries follow China’s lead in emotional-AI governance.
• How infrastructure bottlenecks (power, land, cooling) affect AI deployment costs.
• What frameworks emerge for controlling autonomous agents safely.
• Which AI investments begin to pay for themselves and which remain speculative.

Closing Thoughts

The last week of 2025 felt like more than an end — it felt like a turning point.

AI is still accelerating, but now within visible boundaries: regulatory, physical, and design-oriented. The question for 2026 is not whether AI will get more capable, but how carefully we build and guide those capabilities for real human use.

I’m curious: which of these shifts will matter most for your projects this year?

Happy building — and welcome to 2026.