OpenAI has launched ChatGPT Wallet, a payment layer embedded directly into ChatGPT Work, Codex, and its dots agent platform. Agents can now complete purchases autonomously, but only after explicit user approval. Existing card protections apply.

The architecture matters more than the feature. This is not a chatbot with a buy button. It is a credentialed agent layer where non-human actors execute financial transactions on behalf of users inside a controlled approval loop. The scope of what agents can purchase, and under what conditions approval is triggered, is what readers need to scrutinize in the full piece.

The next question is liability and limits. When an agent misinterprets a purchase instruction, who absorbs the cost? The original post addresses the approval mechanism but the edge cases around autonomous spending behavior are where this gets genuinely consequential. Read it for the architecture details, not the announcement.

[READ ORIGINAL →]