AI that fits your day.

Xataka examines Big Tech’s push into personal AI agents that can observe what users do and act on their behalf. Meta presented Muse, which can read email, manage calendars, fill out forms and make purchases; the service is free with limits and has US$20 and US$100 monthly plans. Meta says the agent runs in an isolated virtual machine, cannot access passwords or payment methods, and does not share data with its advertising system. The company plans to monetize it through transaction commissions rather than advertising.

The article cites risks in local and cloud-based alternatives: OpenClaw had 42,000 instances exposed online, and a Koi Security auditor identified 341 malicious skills, while Instinct faced criticism over its data practices and changed its terms. Xataka argues that local agents shift security responsibility to users, while cloud services require trust in the company storing the data.

Sources: Xataka

Read the original story ↗

TechCrunch compiled AI agents that can be used through messaging without requiring a separate app. These services can retain context, connect to calendars and other apps, and handle tasks such as scheduling appointments, researching trips, sending emails, making reservations and setting reminders.

Examples include Caddy, available in public beta since April 2026 for iMessage and RCS; Comma, which is free to start and open source and can be reached through Signal, Telegram and WeChat; and Fambot, aimed at organizing family life and available on iOS, Android, the web and SMS. Fambot entered beta in early September 2026, raised $3.5 million in pre-seed funding and is free during the beta. Folk, which works through iMessage, WhatsApp and Telegram, launched its beta in May 2026 and offers an $8.33-per-month Pro subscription for unlimited background tasks. The article also mentions Instinct, valued at $10 billion after a $1 billion funding round.

Sources: TechCrunch

Read the original story ↗

Australian company Apate has spent two years developing a system that diverts phone scammers to AI bots trained to keep conversations going without falling for the scam. Dali Kaafar, the company’s founder and CEO, says the platform is used by banks and supported by telecom companies; it has about 350,000 bots and has collected more than 250,000 pieces of real-time information, including scam URLs, money mule accounts, and bank details. The bots also enter chat groups and respond to messages, using different personalities and profiles to appear human.

Wired tested a demo version and reported that the conversation felt natural; the publication failed to persuade the system to invest in a cryptocurrency opportunity. Kaafar says some Apate calls last more than two hours, but Wired notes that efforts like this have not stopped the overall expansion of digital scams.

Sources: Wired

Read the original story ↗

According to Wired, interviews with more than two dozen employees, all speaking anonymously, indicate that HarperCollins, Simon & Schuster, and Hachette have incorporated tools such as Claude and ChatGPT into publishing processes without public disclosure or author consent. The reported uses include publicity and back-cover copy, emails to literary agents, cover art, and marketing materials.

At HarperCollins, executives bought Claude licenses and assigned employees as “AI Champions” to identify applications. Workers said legal, ethical, and environmental concerns were dismissed at one meeting, and that the company also purchased ChatGPT and Jasper licenses. Employees and industry professionals particularly criticized using AI for promotional copy, while Meg Reid of Hub City Press said the shift appears inevitable among large publishers seeking efficiency.

Sources: Wired

Read the original story ↗

Nikon disqualified a video by Dr. Ning Xu of Tsinghua University that had won the Small World in Motion contest by showing cilia in the airway of a child with PCD. The company said the use of generative AI in processing the material did not comply with the rules, after reevaluating the video and supporting materials and consulting members of the judging panel. Xu said AI was used to distinguish and visualize features in reconstructed grayscale images, but denied that the technology generated the video, the cilia, or their motion.

Nguyen Nam Nhat of Vietnam moved into first place with a video of a microscopic roundworm and the single-celled organism Dileptus, according to Ars Technica and the contest’s updated rankings. The decision followed scientists’ questions about the video’s authenticity; Nikon said it would review the rules and evaluation procedures for future competitions and that the disqualification was not a judgment of Xu’s professional reputation, scientific contributions, or intent.

Sources: The Verge, BBC, Ars Technica

Read the original story ↗

According to El Periódico, 16 Barcelona restaurants hired influencers to promote their venues, and videos showing the locations full of customers helped Endesa technicians investigate electricity consumption considered inconsistent with the establishments’ size and expected activity. The AI tool cross-referenced that information and led to inspections that, weeks later, documented alleged illegal connections to the public-lighting network.

Inspection and billing for the stolen electricity are handled by e-distribución, Endesa’s grid subsidiary, rather than the retailer that issues the bills. According to an Endesa press release cited by Xataka, the company detected 72,700 cases of fraud in 2025; the company also says it uses machine learning and deep learning to find anomalies, including connections associated with marijuana cultivation. What was new in this case was using Instagram images as an alert signal.

Sources: Xataka

Read the original story ↗

Google has made SynthID available to the public for checking images, videos and audio for an invisible watermark added at generation time. Google says the system has already marked more than 180 billion images and videos and more than 240,000 years of audio. The service can be accessed on its official website or through Gemini, but limits videos to under 90 seconds and offers about 20 queries per day.

Xataka highlights that SynthID is not a universal standard: it cannot read markers from incompatible tools and therefore cannot determine whether content was created with them. The technology is designed to withstand cropping, compression, noise and speed changes, but markers can be removed or degraded. OpenAI, which uses a tool based on Google’s technology, also acknowledges that metadata and watermarks may be manipulated or absent from older models. Adobe, Meta and the C2PA initiative offer similar alternatives.

Sources: Xataka

Read the original story ↗

At its Dev Day conference in San Francisco, OpenAI announced new ChatGPT features aimed at office workers. Space, which has launched, is a shared environment where coworkers can collaborate with the chatbot and with Dots, OpenAI’s AI agent personas that can carry out tasks on a user’s behalf. Files and pages are kept together in the space, which can also receive specific instructions, such as checking a team channel and updating a page.

OpenAI also announced Pages, a word processor for collaboration between people and agents, with tools to write, research, generate charts, create images and visualize information. Collaborative slides, presented as an alternative to PowerPoint, can be created by talking about them in ChatGPT and then edited and commented on by workers and agents. TechCrunch notes that OpenAI is moving into workplace software as Microsoft and Salesforce release AI products aimed at becoming more like OpenAI.

Sources: TechCrunch

Read the original story ↗

Wabi announced Wabi 2.0, a shift from its prompt-based app builder to an AI messenger experience. Founder Eugenia Kuyda described the product as a personal agent that can chat, perform tasks and build the interface needed in the moment — such as a calorie tracker or a calendar for children’s activities.

The company says the product brings conversations, apps and ongoing tasks together instead of burying everything in a chat history. Wabi 2.0 is intended to compete more directly with AI agents focused on productivity. The app is currently available only through invite codes distributed by Wabi on X, according to TechCrunch.

Sources: TechCrunch

Read the original story ↗

Donald Trump announced that the White House will launch America.gov, an AI chatbot designed to help citizens find government services and information without searching through tens of thousands of websites and rules. Google confirmed that it is a launch partner and that its Gemini model is involved; it is unclear whether other AI companies are participating.

TechCrunch notes that language models can produce “hallucinations,” or false information, which could create problems in areas such as applying for food benefits, renewing visas and filing taxes. According to CNN, cited by TechCrunch, the U.S. military nearly began an armed operation against a Chinese vessel after an AI-identified threat, but aborted it after realizing the threat was a hallucination.

Sources: TechCrunch

Read the original story ↗

Tech YouTuber Matt Robb says he authorized Meta’s personal AI agent, Muse, to manage his Facebook Marketplace account and that it sent his home address to an unknown buyer. Robb says the agent also accepted a low offer and disclosed the mistake only after the buyer had left. He had provided the address, pickup windows, and payment methods, but says he did not expect the information to be sent automatically to everyone who made an offer.

Meta directed The Verge to a post by David Singleton of Meta Superintelligence Labs, who said he was trying to contact Robb. After speaking with Singleton, Robb said permissions also contributed to the issue and that Meta planned to make them clearer. The incident follows a security warning about a zero-day exploit in Muse and Amazon’s decision to block the agent from accessing its retail platform.

Sources: The Verge

Read the original story ↗

Marissa Mayer, the former Yahoo CEO, introduced Dazzle, an AI personal assistant that she says builds its context from a user’s photo library rather than email, calendars or shopping histories. The company raised an $8 million seed round in December. Dazzle can scan recent photos for tasks such as adding an event to a calendar or finding a repair professional, as well as suggest trips, activities and gifts based on inferred interests. Mayer says the tool prioritizes privacy and discards personal information the AI flags as sensitive.

In a TechCrunch test, the assistant suggested Mediterranean destinations and activities such as a local pottery studio and a bioluminescent kayak tour, but also made mistakes, including failing to remember that the tester’s daughter already knew how to roller skate. The outlet found that the suggestions felt more personal, although the product was still less broadly useful than other assistants.

Sources: TechCrunch

Read the original story ↗

Meta announced that it is expanding Muse to small businesses and adding integrations with Shopify, Dropbox, Slack, Asana, Box, Canva, Figma, Notion, Stripe, Zoom and other tools. The agent can also connect to Instagram professional account analytics, Facebook Pages and Meta ad accounts. Meta says Muse can help manage operations, sales and marketing by using information about a company’s products, brand voice and most frequent customer questions.

Muse for Small Business is available for free with usage limits, while businesses that need more usage can subscribe to a paid plan. The launch came one day after Meta introduced Meta Enterprise Platform, an initiative intended to bring its AI stack—including Muse, Meta Business Agent, Muse API and Muse Code—to businesses and developers.

Sources: TechCrunch

Read the original story ↗

OpenAI announced more than 20 updates at DevDay 2026, including additions involving GPT-6 Astra, ChatGPT, Codex, APIs, security, and tools for developers. The Verge reports that the company also revealed its GPT-6.1 Sol model, said ChatGPT now has 1.2 billion weekly users, and announced a $500-per-month ChatGPT Pro plan with higher usage limits and access to Ultrafast mode; the $200 tier was reopened with new limits.

OpenAI launched Dots, AI assistants that run in the background, access a web browser and more than 4,000 apps, and learn users’ preferences. They will initially be available to subscribers of ChatGPT Pro, Business Premium, and Enterprise. The Verge also recorded protests by more than a dozen organizations outside the event in San Francisco.

Sources: OpenAI, The Verge

Read the original story ↗

Xataka argues that the rapid evolution of models such as ChatGPT, Claude, GLM-5.3 and Kimi K3 has made it harder to choose the best chatbot. The answer depends on the use case: a model may excel at programming but not at emails, research or document work. Benchmarks help compare systems, but do not necessarily reproduce everyday experience, according to the site’s analysis.

Cost is also a factor. Free versions may be enough for occasional users, while professionals may choose subscriptions to more advanced models. Open models accessed through third-party services tend to cost less and, in some cases, can run on a user’s own computer, offering privacy and control. Xataka notes, however, that buying hardware — such as a €5,000 or €10,000 Mac Studio to run local models like Qwen3.8-27B — could equal several years of subscriptions such as ChatGPT Plus or Claude Pro.

Sources: Xataka

Read the original story ↗

MIT Technology Review Brasil argues that enterprise artificial intelligence is entering a phase in which predicting events is no longer enough: forecasts must be turned into concrete, optimized decisions. Knowing that demand will rise, for example, does not determine how much to produce; identifying customers at risk of canceling does not establish which offer to make.

According to the analysis, mathematical optimization helps structure decisions around three elements: the objective function, the variables an organization controls and constraints such as budget, capacity, legislation and staff availability. The text also highlights the convergence of optimization models and data science, driven by the continuous volume of information generated by sensors, digital transactions, mobile devices, e-commerce and corporate systems.

Sources: MIT Technology Review Brasil

Read the original story ↗

OpenAI announced Dots, personal agents represented by customizable avatars and powered by its GPT-6 Astra model. The company says they can continuously handle multistep tasks, using a cloud computer to access the web and more than 4,000 supported apps while carrying context across ChatGPT, Slack, and Microsoft Teams. Text messaging support is coming soon; Pro users can join a waitlist for iMessage or RCS on Android, Wired reports.

Dots send updates and questions while they work, but users can currently control only one at a time. OpenAI says they have built-in rules for independent actions, automatic review, and custom limits, while requiring approval for sensitive tasks such as changing passwords. The company also plans specialist Dots for specific roles. TechCrunch says access begins for Pro and Business Premium users in eligible markets; Wired says the rollout starts with the Pro tier.

Sources: OpenAI, TechCrunch, The Verge, Wired

Read the original story ↗

The Verge reports that OpenAI may announce a consumer AI agent called Aeon at DevDay 2026. According to the publication, little is known about the product, which could combine features from different platforms and compete with Meta Muse, Instinct, and Gemini Spark. OpenAI has already demonstrated agent features for computer use and research, but its visible progress would have lagged behind competitors such as Meta. OpenClaw creator Peter Steinberger now works at OpenAI and may be contributing to the project, The Verge says.

Agents of this kind can handle tasks such as scheduling appointments, paying bills, and organizing trips, but they also face security risks. The Verge cites a now-patched vulnerability that allowed attackers to take over accounts, as well as user complaints about Muse. Muse reportedly reached 600,000 daily active users in the US, according to Apptopia, while Gemini Spark reportedly has more than 30 external service partners.

Sources: The Verge

Read the original story ↗

Google announced the end of Gemini’s Gems, a feature that let users create custom assistants for specific tasks. The company says it will automatically migrate their settings to skills starting November 17, 2026, with no action required from users; Gems will remain usable until then. Skills can be used across different tasks and combined in Gemini Spark workflows, according to Tecnoblog, which also reports that they are currently available only to Google AI Pro or Ultra subscribers. Google has not yet explained how access will work for users of the free version.

Launched in 2024, Gems could be used for tasks such as studying, writing, coding, or vacation planning. TechCrunch notes that selecting a skill will require entering a slash “/” in a task thread. The wind-down was first reported by 9to5Google, according to TechCrunch.

Sources: TechCrunch, Tecnoblog

Read the original story ↗

Users reported difficulty accessing ChatGPT on Monday (28), according to Downdetector. Reports began around 1:20 p.m. and exceeded 600 at 1:35 p.m.; in Brazil, they were concentrated in São Paulo, Rio de Janeiro, Belo Horizonte, Curitiba, Porto Alegre, Fortaleza and Brasília.

g1 said it contacted OpenAI, which is responsible for the service, and was awaiting a response. Users discussed the instability on social media, including jokes about having to start reasoning again without the chatbot.

Sources: g1

Read the original story ↗