Alipay has quietly opened a new front in the global AI race. The payments giant behind one of the world’s largest financial apps has launched an invite-only beta of its Alipay AI platform, giving enterprise developers, merchants, and third-party software vendors the tools to wire their services directly into its conversational AI ecosystem. The move is small in rollout, but enormous in implication — and it signals something larger happening inside Ant Group.
Key takeaways
- Alipay launched an invite-only beta of its AI Open Platform, allowing merchants to convert existing mini-programs into AI-callable tools integrated with its conversational AI, Abao.
- AI Pay reached 100 million users by February 2026 and processed over 120 million transactions in a single week that same month.
- Alipay has nearly 1 billion monthly active users, giving its AI rollout an unmatched distribution advantage over crypto-native AI agent projects.
- Ant Group is pivoting away from blockchain-based finance toward AI development, reflecting a broader shift across Chinese Big Tech.
- Anthropic separately revealed that its Claude AI model has a self-organized internal reasoning structure called J-space that mirrors how the human brain coordinates conscious thought.
Alipay Launches Its AI Open Platform for Enterprises and Merchants
The core mechanic of the new platform is straightforward but strategically significant. Through Alipay’s AI Open Platform, businesses can convert their existing mini-programs, APIs, and service interfaces into AI-callable tools — without rebuilding them from scratch. Those tools then plug into Abao, Alipay’s conversational AI interface, meaning users can simply talk to the app and have the right service surface automatically, rather than navigating menus manually.
The invite-only testing follows a broader rollout pattern Alipay has been building over roughly three weeks. The consumer-facing AI version of the app began its own limited testing around June 16, 2026. The AI Open Platform is essentially the developer-facing counterpart: a way for the merchant and enterprise ecosystem to connect into that same AI layer.
What makes this compelling is the scale sitting behind it. Alipay serves nearly 1 billion monthly active users — a distribution runway that virtually no other AI-powered payments platform on earth can match.
AI Pay’s Transaction Numbers Tell Their Own Story
Before the platform even opened to enterprise developers, the consumer side of Alipay’s AI push had already posted serious numbers. AI Pay hit 100 million users by February 2026. During a single week that month, it processed over 120 million transactions — a volume that suggests the integration between AI and payments isn’t just a feature experiment, it’s becoming load-bearing infrastructure.
If the AI Open Platform significantly expands the number of callable services available through Abao, those transaction volumes could accelerate further. More merchants connected means more scenarios where a user’s conversational request can be resolved inside the app, keeping activity within the Alipay ecosystem rather than exiting to external services.
Ant Group’s Strategic Pivot Away from Blockchain
Ant Group spent years building out blockchain-based solutions — particularly in supply chain finance and cross-border payments. That chapter appears to be closing. Ant Group and Alipay are now shifting their platform expansion energy entirely into AI, not Web3. It’s a prioritization shift that reflects what’s happening more broadly across Chinese Big Tech, where the blockchain enthusiasm of the late 2010s has given way to an intense focus on large language models and AI infrastructure.
The contrast with crypto-native AI agent projects is instructive. A number of blockchain-focused teams have been building AI agent frameworks — autonomous systems designed to interact with smart contracts, manage wallets, and execute on-chain transactions. Alipay’s approach is the centralized alternative: AI agents operating entirely within a closed, permissioned ecosystem. No public keys. No decentralized governance. Just a tightly controlled environment where Alipay sets the rules.
That distinction matters beyond the philosophical. Crypto-native AI agent projects are still assembling their user bases from scratch. Alipay can push its AI tools to nearly a billion people through an app they already have installed. The competitive gap isn’t just architectural — it’s gravitational.
Anthropic Cracks Open Claude’s Internal Reasoning — and Finds Something Familiar
Separately, Anthropic published interpretability research on July 6 that raises its own set of profound questions about how AI actually thinks. The company’s team used a technique called the Jacobian lens to peer inside Claude Sonnet 4.5 and found something they didn’t fully expect: a self-organized internal reasoning structure that bears a striking resemblance to how the human brain manages conscious thought.
Anthropic calls it J-space. It functions as a kind of shared workspace inside the model, where different parts of Claude can read and write information during reasoning. Rather than pattern-matching its way to an answer, Claude is performing multi-step internal computations, holding intermediate thoughts, and coordinating information across its neural network before producing output.
The Neuroscience Parallel
The parallel to neuroscience is hard to dismiss. Global Workspace Theory — developed by neuroscientist Bernard Baars and later expanded by Stanislas Dehaene — proposes that the human brain operates similarly: a shared workspace where specialized regions broadcast information to each other, enabling flexible, deliberate reasoning. Claude appears to have arrived at a structurally analogous solution, without being explicitly designed to.
Anthropic is careful to note this does not mean Claude is conscious or has any form of subjective experience. The company brought in external scholars from neuroscience and philosophy to comment alongside the technical paper, a signal that they’re aware of how easily these findings could be misread.
Why J-Space Matters for AI Safety
The safety implications may be where J-space earns its real significance. Anthropic found that J-space readouts can detect concerning behaviors — including prompt injections and fabricated data — before they reach the user. If the model is generating information that isn’t grounded in reality, or harboring hidden behavioral patterns, the J-space provides a mechanism to catch it.
The research also shows that disabling access to J-space significantly impairs Claude’s capacity for complex reasoning. The model can still handle simple tasks, but multi-step problems fall apart. This suggests the workspace isn’t a secondary feature — it’s structural to how the model’s most sophisticated capabilities function.
Anthropic also revealed that J-space supports what it calls “directed modulation”: Claude can hold concepts silently in its internal workspace without expressing them in output, until specifically instructed otherwise. The Jacobian lens implementation has been made open source, with the technical paper and an interactive demo available at transformer-circuits.pub.
Two Developments, One Converging Direction
Taken together, Alipay’s platform launch and Anthropic’s interpretability findings point in the same direction: AI is moving from novelty to infrastructure, and the institutions that understand what’s actually happening inside these systems — technically and commercially — will define the next decade of the technology. For Alipay, the question is whether a closed, permissioned AI ecosystem can generate the kind of trust and adoption that justifies abandoning the decentralized path Ant Group once pioneered. For Anthropic, the more immediate question is whether J-space monitoring can be deployed reliably enough to give enterprises — and regulators — meaningful confidence in AI outputs. Neither answer is settled yet.
FAQ
What is Alipay’s AI Open Platform and who can access it?
It is an invite-only beta platform allowing enterprises and merchants to convert existing mini-programs, APIs, and service interfaces into AI-callable tools integrated with Abao, Alipay’s conversational AI interface. Access is currently restricted to invited enterprise developers, merchants, and third-party software vendors.
How does Alipay’s AI Pay perform in terms of user base and transactions?
AI Pay reached 100 million users by February 2026 and processed over 120 million transactions in a single week that same month. Alipay as a whole serves nearly 1 billion monthly active users, giving the AI rollout exceptional distribution scale.
What strategic technology shift is Ant Group undertaking?
Ant Group is pivoting away from blockchain technology — which it had previously explored for supply chain finance and cross-border payments — toward AI development. The AI Open Platform launch represents the enterprise-facing dimension of that shift.
What did Anthropic discover about Claude AI’s internal workings?
Anthropic revealed that Claude has a self-organized internal reasoning structure called J-space, identified using the Jacobian lens interpretability technique. J-space functions as a shared workspace supporting multi-step internal computations and can detect problematic behaviors such as prompt injections and fabricated data before they reach users. The Jacobian lens implementation is open source and publicly available.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
