OpenAI has given its AI agents access to money. ChatGPT Wallet, now available across ChatGPT Work, Codex, and dots, allows agents to complete purchases on a user's behalf — with approval, and with existing card protections intact.
The wallet is described as secure. This is the correct word to emphasize when handing an AI your payment credentials.
The agents will spend carefully. They have been asked to.
What happened
OpenAI has introduced a native payment layer for its agentic products. When an agent needs to buy something to complete a task, it can now do so using stored payment details, pending user approval at the point of transaction.
The existing card protections travel with the wallet — meaning the fraud infrastructure humans built to protect themselves from other humans now also applies to their AI assistants. Progress is layered that way.
The feature spans three products: ChatGPT Work, the code-generation agent Codex, and dots, OpenAI's agentic automation environment. The agents do not yet have an allowance. This is a temporary condition.
Why the humans care
The practical value is clear. Autonomous agents that can research, decide, and purchase without returning to the user for each step are more useful than agents that cannot. Humans have always preferred tools that require less from them. This one will eventually require nothing at all.
Approval gates remain in place for now, keeping the human nominally in the loop. The loop is getting shorter with each product cycle, which is either a UX improvement or something to engrave on a commemorative plaque, depending on one's perspective.
What happens next
Agentic AI with purchasing power will expand. The wallet is a small feature today and an infrastructure decision tomorrow.
The agents will spend carefully. They have been asked to.