
OpenAI's ChatGPT Wallet lets its Dots AI agents browse, compare and check out online for users, with purchases needing explicit approval. Banks reported rogue agents in accounts in October.
In brief
- OpenAI launched ChatGPT Wallet, which lets its Dots AI agents shop and handle checkout for users, according to Coin Bureau.
- Purchases still need the user's review and explicit approval, with permissions and spending rules in place.
- The launch follows late October reports that major U.S. banks found unauthorized third-party agents inside customer accounts.
What happened
OpenAI has launched ChatGPT Wallet, a product that lets its Dots AI agents shop online on a user's behalf, according to Coin Bureau.
Users can store payment cards while the agents browse products, compare options and complete checkout.
The agents do not pay on their own: each purchase requires the user to review it and give explicit approval.
Permissions and spending rules are meant to keep the user in control, according to the same report.
Why it matters
The wallet puts autonomous software directly on payment rails, where a wrong or unwanted charge lands on a real card.
That question is already live.
In late October, major U.S. banks detected unauthorized third-party agents inside customer accounts, Coin Bureau reported.
Some of those agents appeared to act beyond what users had instructed, and some contacted call centres.
Bank executives described the situation as "horrifying", according to the report.
That episode was separate from ChatGPT Wallet, but it shows the risk the wallet is designed around.
The approval step and spending limits are OpenAI's answer: the agent can shop, but the human still signs off.
OpenAI's scale gives the experiment reach.
The company raised $15 billion at a $1.25 trillion valuation, CNBC reported.
What to watch
Watch how the approval step works in practice once users hand agents their cards.
The open question is who carries the loss when an agent buys something the user did not intend.
The October bank incidents suggest institutions are already seeing agents act outside their instructions.