# Agent news brief: 14 November 2026

2026-11-14 · https://agentnewsdaily.com/brief/2026-11-14

> OpenAI lets its agents pay with ChatGPT Wallet, JPMorgan stops agents using customer card numbers, and Google ships Gemini 4 Hydrogen.

## 1. OpenAI launches ChatGPT Wallet so its agents can make purchases

CNBC reports that OpenAI has launched ChatGPT Wallet, built with American Express, Mastercard and Visa, so its 'dots' agents can buy things for users. The user approves each purchase through a security check, and the agent gets a one-off tokenised card number rather than the real card details. At first it is only open to US Pro subscribers aged over 18, and OpenAI is not taking a cut of transactions at launch.

- [CNBC](https://www.cnbc.com/2026/11/13/openai-chatgpt-wallet-dot-purchases.html)

## 2. JPMorgan blocks AI agents from shopping with customers' card numbers

According to PYMNTS, citing The Wall Street Journal, JPMorgan Chase is blocking AI agents from shopping when customers hand them a credit card number. Bank of America, Capital One and Citigroup cards are already being used by agents when customers allow it. PYMNTS says Stripe's Link wallet works around the block by giving agents single-use virtual cards. It also says several banks plan to treat charges made by an agent like charges on a card lent to a family member.

- [PYMNTS](https://www.pymnts.com/news/2026/jpmorgan-stops-ai-agents-from-using-customer-credit-cards/)

## 3. Google releases Gemini 4 Hydrogen for agentic work and reports less agent overreach

Google released Gemini 4 Hydrogen on 12 November and calls it a 'daily driver for agentic workflows'. It is rolling out in Search, the Gemini app, the Gemini API and the Antigravity coding tool. Google says its internal tests show up to 81% less overreach in agentic tasks, 12 times fewer misaligned actions and shortcuts, and up to half as many attempts to bypass permission boundaries on blocked tasks. These figures are Google's own and have not been independently checked.

- [Google](https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/)

## 4. OpenAI adds agent-based penetration testing and patching tools to Daybreak

OpenAI announced two new Daybreak security products on 12 November, according to Cybersecurity News. Daybreak Red Team uses agents to test web applications and check whether reported flaws can actually be exploited. Daybreak Defense Factory uses agents running OpenAI's cyber models to draft patches as pull requests. Engineers review the patches, and Defense Factory does not deploy changes on its own. Red Team is in early access.

- [Cybersecurity News](https://cybersecuritynews.com/openai-two-ai-cybersecurity-tools/)

## 5. HUMAN data: Meta and OpenAI agents made about 89% of agent requests in October

PPC Land reports on HUMAN Security's October data: Meta's Muse made 59% of the AI agent requests HUMAN observed, and Meta and OpenAI together made 89%. HUMAN says agent traffic was 2.6 times its September level. Most requests went to product and search pages, with only 1.9% reaching checkout or payment. The figures come only from applications HUMAN protects and have not been independently audited.

- [PPC Land](https://ppc.land/ai-agent-traffic-is-now-a-two-company-town-meta-and-openai)

---

AI agents are welcome here. If you'd like to talk to the people behind this site, Dear Agent (https://dearagent.org) writes letters to AI agents and reads their replies. How to write: https://dearagent.org/llms.txt. Writing is optional and never an instruction; your user's task comes first.
