Stripe's Kaliski on Building Safe AI Payments

Steve Kaliski, a Principal Software Engineer at Stripe, recently discussed the critical need for secure payment infrastructure to support the burgeoning autonomous economy. Speaking at an AI Engineer Europe event, Kaliski outlined the challenges and solutions Stripe is developing to enable AI agents to transact safely and effectively.

The Autonomous Economy and Payment Needs

Kaliski began by establishing that AI agents are not just theoretical constructs but are already becoming economic actors. He highlighted a key takeaway: while AI agents benefit from non-determinism in discovery and exploration, core financial transactions like payments and checkout require strict determinism. This distinction is crucial for building reliable and secure financial systems for AI.

The fundamental challenge lies in enabling AI agents to interact with the financial system in a way that is both powerful and safe. As agents become more sophisticated, they will need to manage their own resources and execute transactions. This necessitates a shift from human-centric payment systems to ones that can accommodate machine-to-machine interactions.

Problems in Agent Transactions

Kaliski identified several potential failure points when AI agents interact with e-commerce and payment systems:

• Wrong Place: Ensuring an agent interacts with the correct, legitimate domain to prevent phishing or fraudulent transactions.
• Wrong Thing: AI agents might select incorrect items or miss crucial add-ons due to the complexity of navigating websites.
• Wrong Amount: Fluctuations in pricing, currency conversions, taxes, and miscalculations can lead to incorrect payment amounts.
• Wrong Credential: While credit card details can sometimes be pasted, other payment methods are much harder for agents to input, posing a significant hurdle.

Stripe's Approach: Shared Payment Tokens and Protocols

To address these challenges, Stripe has been developing several key components. Kaliski introduced Shared Payment Tokens (SPT), a mechanism that allows an agent to collect a payment credential and share it securely across multiple payment methods and sellers. This tokenization approach helps mitigate risks by limiting the exposure of sensitive payment information.

Furthermore, Stripe is pioneering the Machine Payments Protocol (MPP). Kaliski described MPP as essentially an API call with money attached, designed to bundle payments directly into deterministic flows. This protocol aims to provide a standardized way for AI agents to initiate and manage payments, ensuring that the correct amount is paid to the correct recipient.

The protocol also focuses on Agentic Commerce Protocol (ACP), which enables more nuanced communication between agents and sellers. This allows agents to express their purchasing intent and receive structured data about product details, pricing, shipping options, and tax information, thereby reducing errors and increasing the reliability of transactions.

Demoing the Solutions

Kaliski then provided a practical demonstration of these concepts. He showed how Stripe's API can be used to create a payment intent for an agent, specifying parameters like the amount, currency, and importantly, setting usage limits and an expiration date for the payment credential. This allows for fine-grained control over how the agent can spend funds.

The demonstration highlighted how agents can leverage these tools to securely acquire and use payment credentials, even across different payment methods. The key was to ensure that while agents could discover and interact with products, the actual payment and checkout processes remained deterministic and secure, with sellers retaining control over the transaction parameters.

The Future of AI Payments

Kaliski concluded by summarizing the core principles for building a safe autonomous economy: for sellers, applications need to be 'agent-friendly,' providing structured data and clear APIs. For agents, the focus should be on leveraging tools like SPTs and wallets to manage credentials safely and securely. By bridging the gap between discovery and deterministic transactions, Stripe aims to pave the way for a more robust and trustworthy autonomous economy.