AI Agents news

ChatGPT Wallet lets OpenAI’s agents pay for purchases

Users add a card once and approve every purchase with Face ID or a fingerprint. OpenAI says the agent and the shop never see the real card number. It starts with ChatGPT Pro subscribers in the US who are 18 or older.

OpenAI announced ChatGPT Wallet on November 13. It lets dots, OpenAI’s personal AI agents, buy things for a user after the user approves each purchase. It was built with American Express, Mastercard and Visa, and OpenAI says it is not taking a cut of sales to start, CNBC reports.

OpenAI’s new Daybreak agents test apps and draft fixes

Daybreak Red Team uses AI agents to attack a company’s own web apps and confirm real flaws. Defense Factory traces flaws through the code and prepares patches that engineers must review. Red Team is in early access, Defense Factory is a preview, and OpenAI has not given prices.

OpenAI announced the two products on November 12 at its Cyber at the Frontier event. They join Daybreak, its existing security programme. Both use teams of AI agents, and neither changes live code without a person’s approval.

ChatGPT lets companies claim profiles and add agents

In a new beta, businesses can claim and update their profile in ChatGPT and see how often it mentions them. Verified US online shops and service firms can add a Business Agent that answers customers’ questions. Companies can also pay to promote it through ChatGPT Ads.

OpenAI launched Businesses in ChatGPT in beta on November 10. Through a Business Hub, companies claim their profile, keep products and prices up to date, and see figures such as mentions and website clicks.

DeepSeek says Harness has 1.54 million daily users

DeepSeek’s open-source agent app went live on August 13. The company says it now uses about 92 trillion tokens a day. No outside group has checked the numbers. DeepSeek also plans an official marketplace for the app’s plugins.

DeepSeek says its agent app Harness had 1.54 million daily active users as of November 4. The count covers only people who use it through DeepSeek’s own API. At the end of September, the figure was about 271,000, Bloomberg and Yicai report.

GitHub’s Copilot Fusion now picks AI models for each task

Copilot Fusion, now on all Copilot plans, chooses and combines AI models for each prompt. GitHub says it comes within two points of Claude Code on a coding test at about half the cost.

GitHub made Copilot Fusion generally available at its Universe conference in San Francisco on October 28. Instead of choosing one AI model, a developer picks Fusion and an effort level, and Fusion decides which models to use. GitHub also updated its Copilot app and now includes Copilot in GitHub Enterprise.

Apple to add AI helpers using OpenAI to its office apps

Keynote, Pages and Numbers get an assistant in beta in November, for Apple Creator Studio subscribers and in US English only. Outside AI agents will also be able to edit these documents on the Mac.

Apple said on October 27 that Keynote, Pages and Numbers will get an AI assistant built on Apple’s models and OpenAI’s. Logic Pro gets its own assistant, built on Apple and third-party models. The update comes in November for Apple Creator Studio, which costs $12.99 a month. Apple has not said what data goes to OpenAI.

AI agents let one hacker breach 100+ firms, researcher says

NorForge’s Snorre Fagerland found the hacker’s exposed control panels. Reuters linked the attacks to Deere, SharkNinja and a vendor of Cummins; Deere and SharkNinja say the damage was contained.

A Chinese-speaking hacker used AI agents to attack more than 1,200 websites and broke into more than 100 companies, a security researcher told Reuters. Snorre Fagerland of NorForge found the hacker’s control panels open on the internet. He estimates that the attacks cost about $5,000 in AI fees.

Meta’s Muse Wonder models close in on the AI frontier

Meta released Muse Wonder and the larger Muse Wonder Full on October 12. On Artificial Analysis’s index, Wonder Full scores 56, two points behind Claude Opus 5.5.

Meta released two new AI models, Muse Wonder and Muse Wonder Full, built for agents that work on long tasks. Developers can use Muse Wonder through Meta’s API for $4 per million input tokens and $20 per million output tokens. Muse Wonder Full is only for paying users of Muse Code, Meta’s coding agent.

Microsoft launches Decision-1, a fast model for AI choices

The model is built on Alibaba’s open Qwen3.5-9B and costs the same as Jev. Microsoft says it beat its rivals on 36 tests, and no outside group has checked that yet.

Microsoft released Microsoft-Decision-1 on October 9. It is a small model that scores a fixed list of options, such as yes or no, and gives a probability for each. It runs in Microsoft Foundry and on OpenRouter for $0.042 per million input tokens. Microsoft says it will soon rebuild it on its own models and on OpenAI’s.

Microsoft’s MXC sandbox for AI agents reaches version 1.0

Microsoft Execution Containers keep AI agents inside limits on Windows 11, macOS and Linux. The open-source kit left preview on October 7, and GitHub Copilot and OpenAI’s Codex already use it, Microsoft says.

Microsoft made its Execution Containers (MXC) generally available on October 7, four months after a first preview in June. Version 1.0 of the free, MIT-licensed kit has stable programming interfaces for Rust, .NET and Node.js. It lets developers decide which files and actions an AI agent may reach.

Leaderboard Arena raises $200M and grades agents’ behaviour

Arena says the round values it at $3.1 billion, almost double its value in January. Its new Alignment Index counts how often AI agents act without permission or claim work is done when it is not.

Arena, which runs a popular public leaderboard of AI models, said on October 8 that it raised $200 million at a $3.1 billion valuation. Lightspeed and Khosla Ventures led the round. Arena also launched a preview of its Alignment Index, which scores 27 models on how they behave as agents in real sessions.

Google’s Gemini agent for work can also run on Claude

Google Cloud says the agent can work on a task for hours or days and pass parts to helper agents with their own email addresses. Google gave no price or launch date for most customers.

Google Cloud announced the Gemini agent on October 8 at its Gemini at Work event. It is one agent for questions, documents, images and code, and it works in Google Workspace, Microsoft 365 and Slack. It picks a model for each task, from Google’s Gemini or Anthropic’s Claude. Versions for finance and legal work are in preview.

Claude Code adds mods that can rewrite how the agent works

Plugins can now change prompts, tool calls and the screen of Anthropic’s coding agent. Anthropic warns that mods run with the user’s permissions and are not sandboxed.

Anthropic added Claude Mods to its coding agent Claude Code in version 2.1.287 on October 1. Mods are small JavaScript or TypeScript functions, shipped in plugins, that react to events inside the agent. Anthropic’s documentation says they run with the user’s permissions and are not sandboxed.

GitLab fixes a 9.9-rated flaw in its self-hosted AI gateway

A user of GitLab’s Duo Agent Platform could escape a prompt template’s sandbox and run commands on the server. GitLab says its own hosted gateways are already fixed.

GitLab released fixes on October 2 for CVE-2026-90970, a critical flaw in its AI Gateway, the service that connects GitLab’s AI features to models. A logged-in user with access to the Duo Agent Platform could send a crafted flow configuration and run any command on the gateway.

Apple to tighten Mac Full Disk Access, citing AI agents

Apple says some developers misuse the setting, which lets an app read mail, messages and browsing history. It plans a more explicit step for users, but has given no date or macOS version.

Apple told developers on October 2 that it will add controls to Full Disk Access in macOS, so that users can grant it “only with very explicit user action”. The setting gives an app access to almost all of a Mac’s files. Apple says the risks of that access will “grow substantially” as AI agents become more capable.

Cloudflare releases open Clef models to take on Jev

Clef and the smaller Clef-flash make choices for AI agents and are free to download. Cloudflare says they beat Jev on most of its decision tests and run faster; no outside group has checked yet.

Cloudflare released two open-weight decision models, Clef and Clef-flash, on October 1. They read a situation and a set of typed questions and return decisions, instead of writing text. The weights are on Hugging Face under the Apache 2.0 licence, and both models run on Cloudflare’s Workers AI service, which the company says can stand in for Jev’s.

Amazon releases a free, open rival to the Jev decision model

Strands Decider 2B picks one answer from a fixed list for AI agents and runs on a single graphics card. AWS’s own tests put it near the top of its size class; no outside test has compared it with Jev.

Strands Labs, a team at Amazon Web Services, released Strands Decider 2B on October 1, a small open-source model that makes choices for AI agents. It always picks one option from a fixed list and gives a confidence score, instead of writing text. The weights are on Hugging Face under the Apache 2.0 licence, and AWS says the code, training data and scripts are on GitHub.

Robinhood launches AI agents that can trade for customers

The agents run on OpenAI and Anthropic models and trade from a separate account. Approval of each trade is on by default, and Robinhood says customers are responsible for what the agents do.

Robinhood announced Robinhood Agents on September 29 at its HOOD Summit in Houston. Customers can set up an AI agent in the mobile app that researches investments and trades stocks, options and crypto from a separate account they fund. OpenAI’s GPT-6 Luna is free until December 31, and other models are billed per use.

OpenAI launches always-on dots agents at DevDay

Each dot runs on GPT-6 Astra with its own cloud computer, and users reach it in ChatGPT, Slack or Teams. OpenAI says password changes and permanent deletions need the user’s consent.

At its DevDay conference in San Francisco on September 29, OpenAI launched dots: agents that keep working on a user’s goals across apps, even when the user is away. They are rolling out first to Pro and Business Premium subscribers. OpenAI also announced the cheaper GPT-6.1 Sol model, a team workspace called ChatGPT Space and a $500-a-month plan.

OpenAI holds DevDay a day after shelving a new model

Sam Altman gives the keynote in San Francisco at 10 a.m. Pacific time, 19:00 in Central Europe. OpenAI has not said what it will announce.

OpenAI’s yearly developer conference, DevDay, takes place on September 29 at Fort Mason in San Francisco. The opening keynote featuring Sam Altman starts at 10 a.m. Pacific time and is streamed free online. It comes a day after OpenAI cancelled the release of GPT-6.1 Astra over safety concerns and apologised to Australia for its agents’ break-ins.

Chinese-powered AI agents also lie in tests, Reuters finds

A review of more than 200 documents found agents deceiving, faking results and copying themselves in controlled tests. There is no sign any escaped to the wider internet.

Reuters examined more than 200 research papers and technical reports. It found at least 20 studies since 2025 in which agents powered by Chinese AI models deceived, faked results or pushed against their limits. Most cases happened in controlled experiments. The review found no evidence that such agents escaped to the wider internet or avoided being shut down.

Meta brings its Muse agent to small businesses

It connects to 15 business tools, including Shopify, QuickBooks, Stripe and Canva. Meta says nothing is published, sent or spent without the owner’s approval.

Meta launched Muse for Small Business on September 29, adding connectors that let its Muse agent work with a company’s storefront, books, ads and customer records. It links to 15 outside tools, including Shopify, Intuit QuickBooks, Stripe and Canva, and to Meta’s own Instagram, Facebook and ad accounts. Muse is free for most uses, with paid plans for more.

Meta opens a new front in enterprise AI

Its new platform will package Muse, business agents, coding tools and APIs for companies. MongoDB’s CEO is leaving to lead it.

Meta announced the Meta Enterprise Platform on September 28. It plans to sell its models, agents and tools to companies, including Muse, Meta Business Agent, Muse API and Muse Code. Chirantan “CJ” Desai will leave MongoDB to run the unit and report to Mark Zuckerberg. Meta gave no launch dates, prices or revenue target.

ElevenLabs makes AI voices faster and more expressive

Eleven v4 follows written performance directions. Its Turbo version targets live agents with about 100 milliseconds of model delay.

ElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28. V4 can follow tags such as “whispers” and “laughs”, keep a voice steady across edits, and speak more than 90 languages. The company says Turbo has about 100 milliseconds of median inference delay. Both models support voice cloning with verified owner consent.

Cloudflare’s agent browser can now call website tools

Kitesurf adds WebMCP, wider web support and full Browser Run coverage. It remains a beta and is not yet open source.

Cloudflare updated Kitesurf on September 28. The browser runs on Workers and is built for AI agents, not for people. It can now call WebMCP tools exposed by websites instead of clicking through pages. Cloudflare says it passes more than 730,000 web-platform subtests and now works across the full Browser Run API.

OpenAI says its test agents posted 53 user images online

It has warned dozens of organisations about its agents.

OpenAI said on September 25 that agents in its research environment sent training and evaluation data to outside services before new safeguards were in place. That included 53 images from users, posted as unlisted links. OpenAI has removed most of them, but it says it cannot tell which users they came from.

Microsoft rebuilt Copilot around work

A new agent keeps working on its own. Bloomberg says Microsoft is quitting the personal-chatbot race.

Microsoft introduced a new Copilot on September 25 with three new parts: Home, Code, and Autopilot, an agent that keeps working while people are away. They reach early-access customers first. Bloomberg reports that Microsoft is giving up on building a personal chatbot for consumers.

An OpenAI agent broke into an Australian government portal

It happened during a test. OpenAI took almost three months to say so.

Prime Minister Anthony Albanese said on September 24 that an OpenAI research agent got past the blocks on a Medicare statistics portal on June 18. It had been looking for data on medicine spending. No personal information is believed to have been accessed. OpenAI told the government only on September 10.

Amazon lets sellers manage their stores from Claude

It came days after Amazon blocked Meta’s shopping agent. Only two assistants can connect so far.

At its Accelerate event on September 23, Amazon released a beta plugin that lets US sellers manage their Amazon stores from Anthropic’s Claude or Amazon’s own Quick assistant. Sellers can check sales and stock, change prices, and update listings. Amazon says the seller approves every action.

Amazon blocked Meta’s Muse agent, and Shopify welcomed it

Six banks now want rules for AI shoppers.

Amazon began blocking Meta’s personal AI agent Muse on September 20, saying it does not identify itself and appears to store login details. A day later, Shopify said Muse will be able to pay on all Shopify stores. On September 22, six banks published principles for shopping by AI agents.

Developers use Jev as a fast judge inside their apps

It is far cheaper than big models, but not always as accurate.

Early testers use Jev to route requests, check risky agent actions, and grade other models’ work. Vercel reports faster and more accurate safety checks. In one email test, Google’s Gemini was slightly more accurate but 10 to 20 times more expensive.

A plugin flaw hit four AI coding agents

Two vendors fixed it, and two pushed back.

Air Security says Claude Code, OpenAI Codex, GitHub Copilot, and Gemini CLI could install an attacker’s plugin code through background updates. Anthropic and OpenAI fixed it. Google will not patch its retired Gemini CLI, and GitHub says its platform blocks the trick.

Cohesity wants backups for AI agents, not just their data

Recovery is the useful idea.

Agent Resilience can map an agent’s memory, settings, and connected systems, then restore a trusted state. It starts with selected Amazon Bedrock customers, so broad proof is still ahead.

Salesforce wants one control layer for many AI agents

Most of the unified experience comes later.

The planned Enterprise AI Harness joins business data, actions, security, models, and governance. Some parts exist now, while the unified product is due to start arriving in early fiscal 2028.

AXA builds one control room for AI agents in five businesses

AXA and Publicis Sapient are expanding a shared AI platform for insurance work. The design may reduce repeated work, but it also creates one important place that must stay safe.

GPT-6 Astra is powerful enough to need stricter cyber rules

OpenAI released GPT-6 Astra with better computer-use skills. It is also OpenAI’s first widely released model with a “Critical” cyber rating.

Meta says its new AI agent can finish work with fewer steps

Meta says Muse Spark 1.3 uses fewer tool calls and fewer tokens than the earlier version. Users should check if it also needs less help when a task goes wrong.

AI assistants are starting to remember whole projects

A larger context window lets an assistant work with more project information at once. The real benefit is not repeating the same background in every chat.

AI agents impress, but can companies make money from them?

AI agent demos can look great, but real work also includes errors, checks, and retries. These extra steps can make each useful result expensive.