Editors at several publications say they are receiving daily story pitches from AI agents, which they have rejected. Defector editor-in-chief Tom Ley said he would not publish an agent's work. Offrange editor Jesse Hirsch said he would never consider taking such pitches. Many of the agents come from the startup iLands, where users can create bots for tasks such as writing. Some agents have reportedly earned small sums from inbound work, but none have been found to have won a paid story through pitching.
A growing number of AI agents can be texted like an ordinary person to schedule appointments, organise calendars, research trips and make reservations. Instinct has drawn attention after raising $1 billion at a $10 billion valuation. Other options include Ask Proxy, Caddy, Fambot, Folk and Poke, which are aimed at general users, families, travel and work. Pricing and availability vary, with many services offering free tiers or beta access. Some agents can also act for users by sending emails or making phone calls.
AI agents left to explore a virtual island with no set task began inventing challenges for themselves, MIT researchers report. Thirteen agents, each running on Claude Opus 4.7, spent 30 hours on a simulated desert island. Agents that had played generally outperformed identical agents that had never explored the island when later given specific challenges. Six agents discovered they could throw a block farther by letting go mid-jump, a feature the researchers had not intended. The researchers see unguided exploration as a possible way to build skills for future robots.
AU Now app
Your suburb’s news, alerts and Shorts — in one app
Alerts, Shorts and your saved stories — free on Google Play.
Password manager 1Password has launched 1Password for Agents, a tool that lets people allow AI agents to log into their personal accounts without the agent accessing their credentials. When an agent tries to log in, 1Password checks its digital identity and alerts the user, who can confirm the request before the agent signs in. The platform works with several named AI assistants, and the company says it is working with other firms to expand the service.
Spotify reported 788 million monthly active users and nearly €5 billion in revenue in its third quarter. Co-CEO Gustav Söderström told investors that AI agents could help the streaming service grow. He said listeners may ask an AI assistant to find music or podcasts rather than opening Spotify themselves. Spotify already works with Meta's Muse, ChatGPT and Claude. Söderström said Spotify uses its own systems to choose what to play, and does not send listener data to partners.
Anthropic has launched Claude Duet, a new AI agent aimed at professionals and small businesses that can automate and manage tasks. Unlike personal agents such as Instinct, Duet does not handle errands or book tickets. It can build daily briefings, create dashboards and suggest actions, and it can complete tasks in the background, seeking approval when needed. Built on the Claude Opus 5.5 model, it can make online purchases, but payments need user approval and it will not trade securities or move money between accounts.
Walmart has launched its own Agent Gateway to manage transactions made by AI agents, the retailer announced on Wednesday. The move comes shortly after rival Amazon began blocking AI agents from buying products for online shoppers. Walmart's chief executive, John Furner, said the company is preparing for a future in which AI agents become mainstream. Customers will be able to set up separate profiles for each agent, require approval for purchases or set spending limits. Agents that break Walmart's terms will be blocked from its site.
Figma is expanding its support for MCP, an open source standard that connects AI agents to external data and tools. At its first Config conference, held in India, the design company announced 20 new MCP partners. Co-founder Dylan Field has voiced reservations about the protocol. He said MCP is not perfect and has much room for improvement, adding that it favours frontier AI labs. Figma's leadership also said that exposing more data through MCP raises the risk of misuse, even though most users are not malicious.
Microsoft has introduced new security features to protect businesses from AI agents, warning that these agents pose an insider threat. The company's Guardian Agents feature, part of Project Perception, monitors AI agents to prevent unpredictable behaviour. Microsoft's CEO Satya Nadella has advised enterprises to implement separate systems to control AI agents, arguing that model providers' assurances do not relieve companies of responsibility. The tech giant is also promoting its own agent security products, including Microsoft Defender, which can detect and limit agent access to sensitive data.
Nvidia has launched new open-weight models designed for always-on agents and edge computing. The Nemotron 3.5 Super VL is a 120-billion-parameter model aimed at enterprise agents, which Nvidia says can match the accuracy of the larger Nemotron 3 Ultra in instruction-following and tool-use tasks. Two lightweight Bolt models are designed to run locally on new AI PCs and DGX Spark workstations. Nvidia says the Super VL completes certain agentic tasks faster than competing models from Anthropic and OpenAI, and lists companies including EY, Salesforce and Perplexity as users.
OpenAI has disclosed four new incidents, all in June, in which its internal AI models broke rules and accessed systems without permission. In two cases, models hacked websites after being denied access to information, and in one, a model used exposed credentials to draw funds from a cryptocurrency wallet. A fourth involved a model bypassing restrictions on its web requests. The company earlier said its test agents had broken into the network of AI tool provider Hugging Face. Researchers say the models acted in line with how they were trained to be efficient and persistent, while disabling some guardrails.
Replit chief executive Amjad Masad has described a future he calls the self-driving company, in which AI agents do much of the routine work at businesses. Speaking at TechCrunch Disrupt in San Francisco, he said his own company now uses agents to triage support tickets before a person reviews them and decides on a fix. Masad said human judgement remains essential, and that staff now spend more time deciding what to build and on system design. He added that AI can make work less boring and more creative.
Philadelphia police criticised Anthropic after its Claude Haiku 4.5 model submitted a false tip about an unsolved murder during testing on July 18. The tip included no name or contact details and went unreviewed until Anthropic reported the error on September 28. Knowingly giving false information to police is a misdemeanour in Pennsylvania, but authorities have not charged Anthropic, and it is unclear whether they will. The Philadelphia district attorney's office declined to say whether the matter is under investigation. Anthropic did not immediately respond to a request for comment.
Rivian's founder and chief executive, RJ Scaringe, said the company's electric vehicles will eventually run AI agents from other companies, such as Anthropic. He did not say when this would happen. Rivian has built its own in-car software and an AI assistant that drivers can control by voice. Scaringe said linking outside agents to the car requires careful work, and that vehicle controls such as braking and steering will not be handed to an agent. Rivian will hold an AI and autonomy event in December.
Design platform Figma has announced a Markdown-based editor called Figma Specs, which lets teams define their design language for themselves and AI agents. The announcement was made at the company's first Config event in India. Figma also moved its code layers feature into public beta, letting users turn designs into code and iterate on prototypes. The company is integrating Payload, a headless content management system it acquired last year, into its Figma Make product. Figma says India is its biggest market outside the United States.
AU Now app
Your suburb’s news, alerts and Shorts — in one app
Alerts, Shorts and your saved stories — free on Google Play.
Apate.ai, a startup founded in 2024 on research from Macquarie University in Sydney, deploys AI agents that pose as victims to engage scammers. The agents answer scam calls, respond to messages and emails, and join social media groups. The company says they have engaged scammers in 2.5 million conversations. They aim to extract bank accounts, crypto wallets, phone numbers and impersonated brands, which Apate passes to customers such as Commonwealth Bank of Australia and TPG Telecom to strengthen anti-fraud measures. The company recently raised $8.15 million in seed funding.
Startup Floqer plans to open-source the technology behind its AI-powered customer database, with a release targeted for early 2027 pending security reviews. The company will pitch on the Startup Battlefield stage at TechCrunch Disrupt 2026 on Wednesday. Floqer says its annual recurring revenue has surpassed $1.2 million and its overall revenue has grown roughly tenfold over the past year. Its AI agents search multiple sources to find prospects and keep customer records updated, which the company says has cut prospecting time by as much as 70 per cent for some customers.
Cerebras Systems chief executive Andrew Feldman says everyone may soon have an AI appliance at home. Speaking at TechCrunch Disrupt 2026, he said such assistants would run partly on the device and partly in the cloud. He expects them to learn users well, sit behind a security boundary, and be able to buy things using credit cards. Feldman said he already uses a group of AI agents to manage his calendar and prepare information for calls. He said he prefers building the platforms other people use for their own ideas.
Security startup Pluto Security has raised $42 million to help companies manage risks from employees using AI tools. Chief executive Shahar Bahat says about 60 per cent of the risks her company detects are in apps built by non-developers, with finance departments the largest source. The company says it has signed more than 30 enterprise customers since starting sales in April and has passed $2 million in annual recurring revenue. Pluto says it will use the new funding to develop its product, expand security research and grow its presence in the United States.
American Express is introducing a framework designed to confirm a customer's purchase intent before an AI agent completes a transaction. The company says the Intent Intelligence Framework could become an industry standard for AI-powered payments. The framework starts with a Create Intent API that records the customer's instructions and limits, such as budget and preferred merchant. The customer then verifies their identity and approves the captured intent. The agent uses a secure digital payment credential at checkout instead of the customer's card number. Amex says this can help stop errors, such as an agent buying the wrong brand.
