Realtime AI News
OpenAI launches ChatGPT Wallet, letting agents pay only with user approval
OpenAI has introduced ChatGPT Wallet, a payment feature that lets agents in ChatGPT Work, Codex and dots complete purchases with user approval while keeping existing card protections in place. TechCrunch reports the rollout starts as a limited release for U.S. Pro subscribers, with agents able to shop and check out on a user's behalf.

OpenAI introduced ChatGPT Wallet on November 13, describing it as a secure way for agents inside ChatGPT Work, Codex and dots to complete purchases with user approval while existing card protections stay in place.
Two phrases carry most of the announcement's weight: user approval and existing card protections. The first means an agent cannot spend on its own authority and each transaction needs the user's confirmation. The second places the flow inside the banking system's familiar security and dispute framework — OpenAI's wording does not present this as a new payment network.
TechCrunch adds the rollout detail: the wallet begins as a limited release for U.S. Pro subscribers, with agents able to shop and check out on a user's behalf. Both sources agree on the division of labor — the agent executes the purchase, the human authorizes it.
The wallet is wired into entry points that already exist. ChatGPT Work handles task execution inside workstreams, Codex serves coding and development work, and dots is the third product line named in OpenAI's description. Attaching payments to those surfaces gives action-taking agents the one capability they were missing.
The real design question is scope of permission, not the checkout button. Once an agent can place an order, a mistake costs money rather than credibility, so approval granularity, an auditable trail and reversibility matter more than which payment rail carries the transaction. OpenAI leading with user approval suggests it understands that trust is the gating factor for the whole feature.
Several things remain undisclosed in the material available: which merchants are supported, whether limits apply per transaction or in aggregate, how refunds and disputes are assigned, and when the wallet moves beyond U.S. Pro subscribers. Those details decide whether this is a demo-grade capability or a usable checkout channel.
Three things are worth watching: the first merchant mix, the actual approval interaction users will see, and how liability is split among OpenAI, the merchant and the card issuer when the first disputed purchase arrives. For teams building agent products, the signal is that the payment authorization layer is becoming something you either build or integrate.
Why it matters
The move pushes agents from suggesting actions to executing transactions, and makes authorized payments a likely standard layer in agent products. How trustworthy the approval step feels to users will decide how fast this escapes a limited test.
Nearby Updates
All11/13, 22:05
llama.cpp Ships v1.0.0, Calling It a Stronger Foundation for Local AI
The llama.cpp project published a Hugging Face blog post introducing version 1.0.0 under the title “A stronger foundation for the local AI ecosystem.” The version jump is a signal to the toolchains and products that depend on the project's inference engine and GGUF format.
11/13, 23:57
New York opens RAISE portal, requiring large AI developers to register by January 1
New York has launched a RAISE registration portal that requires large AI developers to sign up by January 1, according to a report by Pasquale Pillitteri. The portal turns the state's AI rules into a dated filing obligation, though thresholds and penalties are not detailed in the report.
11/14, 00:00
ChatGPT adds a Wallet so AI agents can pay at checkout
OpenAI has added a Wallet to ChatGPT that lets AI agents shop and check out on a user's behalf, starting with a limited rollout to U.S. Pro subscribers. It moves the assistant past recommendations and into live payments, putting authorization and liability at the centre of agentic commerce.
11/14, 00:03
Gemini 4 Hydrogen costs 76% less per task than 3.8 Flash
An analysis by FourWeekMBA puts Gemini 4 Hydrogen 76% cheaper per completed task than Gemini 3.8 Flash. For agentic workloads that make many calls per job, cost per task says more about a model generation than a benchmark score does.