Stripe Releases 288 New Features for AI Commerce — Including Agent Payments and Real-Time Fraud for Autonomous Transactions

The Facts

Stripe has announced one of its largest single product releases, covering 288 new products and features across its payments platform. AI commerce is one of the release's headline themes — the infrastructure required to handle transactions initiated by AI agents, autonomous software systems, and agentic workflows rather than human-initiated payment flows.

The release comes as the payments industry broadly responds to the emergence of agentic AI systems that need to execute purchases, manage subscriptions, and handle billing events autonomously. Visa's Agentic Ready programme in Singapore (covered separately this week) addresses the same infrastructure challenge from the issuing bank side; Stripe's 288-feature release addresses it from the merchant and developer side.

Alongside Stripe's release, two other significant agentic commerce developments are shaping the regional landscape. Ant International — the global payments arm of Ant Group — announced expansion of its agentic commerce capabilities, with its network now connecting more than 150 million merchants globally with over two billion user accounts. Moomoo, the investment platform popular with retail investors in Singapore, launched a new API feature enabling agentic investing — where AI models can execute investment actions on behalf of users within defined parameters.

Technical Deep-Dive

The core technical challenge Stripe is solving with its agentic commerce features is authentication and authorisation in a non-human payment context. Standard card payment authentication relies on a human reviewing and approving a transaction. Agentic payments require a different model: a user grants an AI agent a scoped payment credential with defined limits (merchant category restrictions, spending caps, time-bounded access), and the agent executes transactions within those bounds without requiring per-transaction human approval.

Fraud detection for agentic payments is also qualitatively different. Traditional fraud models are trained on human behavioural patterns — purchase velocity, geographic consistency, merchant category sequence. AI agents will exhibit very different patterns: purchasing from multiple merchant categories rapidly in service of a user goal, at hours unusual for the user's time zone, potentially across multiple jurisdictions. Stripe's new fraud controls need to be calibrated to distinguish between legitimate agentic behaviour and compromised agent credentials.

For ASEAN developers, Stripe's API-first approach means these capabilities are accessible without requiring enterprise contracts — startups building agentic applications can integrate payment infrastructure that handles AI-initiated transactions through the same developer-facing API they use today.