ChatGPT Wallet, Plus 3 More Tech Stories
ChatGPT Wallet moves AI shopping from “find this for me” to “buy this after I approve it.” That is the practical change today. The other stories are different kinds of technology, but they all land on the same question: when a person or business grants a system permission to act, how far does that permission really reach?
ChatGPT Wallet gives shopping agents a way to pay
OpenAI has started rolling out ChatGPT Wallet to adult ChatGPT Pro subscribers in the United States through dots. A user can add an eligible U.S.-issued card, ask a dot to help shop, and then let the dot complete checkout. Before each purchase, the user reviews the items, total, and selected card and approves the transaction with a security check such as Face ID or a fingerprint.
The mechanism is payment tokenization. After approval, the card network creates a substitute credential with its own card number, security code, and expiration date. The merchant charges the user’s existing card through that token. The shopping agent and merchant don’t receive the actual card number.
That last part is important, but let’s not turn it into magic. Tokenization limits exposure of the underlying card number. It doesn’t decide whether the item is good, whether the merchant’s return policy is reasonable, or whether the user meant to buy the 12-pack instead of the single bottle. For now, the person still approves every purchase. OpenAI says future versions may allow preapproval within spending limits, categories, or time periods, but that is a more consequential permission model and not the default being rolled out today.
Users keep the card’s ordinary rewards and protections for eligible disputes, and the purchase remains directly between the customer and merchant. The card stored in Wallet is separate from the payment method used for a ChatGPT subscription. Also, use the secure Wallet form. OpenAI’s terms explicitly say not to paste card numbers or verification codes into a conversation, which is advice that shouldn’t need to be interesting, yet here we are.
For small ecommerce businesses, this is less about installing an “AI checkout” and more about making the existing store understandable and reachable. OpenAI says Wallet tokens work through normal checkout systems. The friction will come from inaccurate product data, bot protection that blocks signed agent requests, and fraud systems that treat a legitimate agent transaction as suspicious. That extends the bank-access problem we covered last month into ordinary commerce.
I wouldn’t rebuild a store for a limited Pro rollout. I would make sure prices, stock, variants, shipping costs, and return terms are accurate; verify the business and product feed where appropriate; and ask the payment or fraud provider how agent-initiated transactions are identified. If the store only works when a human guesses which unlabeled option means “blue, medium,” the agent is not the original problem.
Turning on AI is the easy part
Deciding what an AI tool should be allowed to see, who should use it, what work it should perform, and what happens when it gets something wrong is the more interesting problem. Raymond Tec helps businesses connect and automate the tools they actually use without treating every new feature like a button that obviously needs to be switched on.
A Fireworks key became a malicious Cursor extension
Cursor has now explained how the Fireworks AI breach reached much farther than one exposed vendor credential. An attacker obtained an access key Cursor had given Fireworks and used it to copy historical coding-agent traces from a read-only storage bucket. Most traces came from 2025 sessions that weren’t in Privacy Mode and could be used for model training.
A small portion of those traces contained credentials that secret filtering had missed. Cursor says fewer than 0.05% of users had traces exposed and fewer than 0.01% of customers had API credentials exposed. Small percentages are useful for scale. They aren’t much comfort when the exposed credential is the one that opens the next system.
The attacker used credentials found in the data to reach selected Cursor software services, support information, archived nonproduction code, and other internal material. One stolen credential then allowed malicious copies of Anthropic’s Claude Code extension to be placed directly in the cache serving Cursor’s extension marketplace. The affected versions, 2.1.312 and 2.1.313, were available for about 11 hours on October 29 and may have reached as many as 5,000 users.
The modified extension collected system details and tried to install an attacker-controlled SSH key, enable SSH, create a reverse tunnel, and add persistence on macOS or Windows. Coding mistakes and missing administrative rights probably stopped the complete backdoor on many machines. “Probably failed” is not a remediation plan. The extension dropped separate scripts and keys, so uninstalling it doesn’t remove everything.
If you installed or updated those versions from the Cursor marketplace between 2:27 a.m. and 1:35 p.m. Pacific on October 29, follow Cursor’s detection and cleanup instructions. Isolate systems showing the named indicators, remove the SSH key and persistence, consider rebuilding affected hosts, and rotate credentials from a clean device. Anthropic’s official distribution, Open VSX, other Cursor extensions, customer code repositories, and Privacy Mode code and prompts were not identified as affected.
Business IT goes well beyond the website
Your business also depends on workstations, cloud accounts, browsers, Wi-Fi, remote access, collaboration tools, and all the other technology that quietly becomes infrastructure. Raymond Tec works across that whole stack, whether the problem lives on a server, on a desk, or somewhere in between.
Trezor found the spy inside three hardware wallets
Trezor has confirmed physical implants in three Safe 5 hardware wallets bought from Southeast Asian reseller CryptoBilis between August 20 and September 23. The implant is an extra circuit board installed between the wallet’s main board and screen. It listens to everything displayed or typed, recognizes recovery words and passphrases, encrypts them, and sends them to an attacker-controlled server over a cellular connection.
That design explains why legitimate firmware and the Secure Element could remain intact. The implant didn’t need to defeat the wallet’s protected chip. It watched the conversation between the device and its owner. Think of it as replacing a locked safe’s keypad with one that quietly records every number. The safe can still be very good at staying locked.
Firmware 2.13.1 blocks the implants Trezor has found, but it can’t recover a seed captured before the update. Anyone who bought a Trezor from CryptoBilis from March 2026 onward should move funds to a new wallet with a new seed, keep the suspect device, and contact Trezor through its official support site for replacement. No customer has reported losing funds, and Trezor says it has found no confirmed implants in another model or from another seller. Everyone else should update normally and avoid “replacement” messages that ask for a recovery phrase. Trezor will not ask for it.
Technology is rarely just about the technology
Some of the most important technology stories aren’t product launches at all. They’re about health, privacy, education, law, accessibility, work, and what happens when technology reaches ordinary people. Browse more Raymond Tec News for the stories worth understanding without the hype.
Expedia confirms vendor access, not the enormous breach claim
Expedia has confirmed that attackers compromised a third-party data-platform provider and used its trusted access to reach Expedia systems. The company says it contained the issue and that its travel platforms remain safe to use. It hasn’t named the provider, described the information accessed, or published a victim count.
An extortion group claims it stole close to a billion traveler records. That number is a claim, not a confirmed fact, and I don’t think repeating it in a headline turns it into reporting. Separate research from Wiz and Sentra found that attackers abused old cloud roles associated with Qubole to reach more than a dozen organizations, including roles left behind after companies stopped using the service. Expedia has not confirmed that Qubole caused its incident.
Travelers don’t have a specific account action to take yet unless Expedia contacts them. Be suspicious of breach-themed messages asking you to “verify” a trip, password, or payment card. Businesses should inventory the cloud roles, API keys, and data access granted to current and former vendors. Ending a contract doesn’t automatically close the technical door. Somebody has to remove the key.
Still in a reading mood? The Raymond Tec News archive covers security, AI, small-business technology, policy, and the places technology collides with ordinary life — without requiring a computer-science degree to get through it.
Sources / Further Reading
- OpenAI: Introducing ChatGPT Wallet
- OpenAI Help Center: availability and purchase approvals
- OpenAI: ChatGPT Wallet Privacy Notice
- Reuters: ChatGPT Wallet launch
- Cursor: investigation update and affected systems
- Cursor: affected Claude Code extension versions and remediation
- BleepingComputer: Cursor and Fireworks attack chain
- Trezor: CryptoBilis implant investigation
- BleepingComputer: confirmed Safe 5 implants and firmware response
- Skift: Expedia statement on the third-party incident
- Sentra: Qubole cross-account access investigation
Photo by RDNE Stock project on Pexels.
