AI that fits your day.

Coyotec, an artificial intelligence radio cohost from Entravision, debuted on a solo weekend program in September 2025 and now appears on Al Aire y Sin Permiso alongside broadcaster Geraldine “GeeGee” Guzman. The avatar speaks Spanish and English, uses informal language, and posts about immigration, racism, and understanding one’s rights when facing ICE.

The pairing’s popularity led the station to move the show to weekday mornings from 8 to 10 a.m. and, as of August, syndicate it in five new markets across California, Nevada, and Texas. According to Entravision audience data cited by Wired, the program saw significant listener growth during its first two months of syndication, although the publication did not provide figures.

Sources: Wired

Read the original story ↗

Three months after Elon Musk announced that Grok would release a full-length, “historically accurate” version of “The Odyssey,” the system published seven minutes, now presented as “Episode 1” of a series. The project was created in collaboration with Wonder Studios and director Dev Walker.

TechCrunch found the footage more convincing visually than other videos recently posted by Musk, with moments that could resemble real actors being filmed. The site’s review said, however, that the clip lacks artistry, inspiration, or human interest: extended narration gives way to bearded faces delivering stilted lines, and the whole production provides little momentum or reason to care about its characters.

Sources: TechCrunch

Read the original story ↗

g1 classified as fake a photo shared on social media and WhatsApp that allegedly shows a protester tearing an Our Lady of Aparecida banner during a rally supporting Flávio Bolsonaro in Belo Horizonte. The image carried the g1 logo, but the outlet says it never published the photo on its website or social media accounts.

According to the Fato ou Fake team, two detection tools indicated the use of artificial intelligence: SynthID identified signals consistent with content generated by OpenAI tools, while Sightengine estimated a 96% probability that the image was created with AI. The content circulated with indignant captions after the rally held on Saturday (24).

Sources: g1

Read the original story ↗

According to The Verge, the BBC's Blue Planet III is keeping less-than-perfect shots to present a more authentic, organic version of the oceans at a time when overly polished scenes can look AI-generated. Executive producer Elizabeth White told The Guardian that the team might previously have cut shots with too much plankton, reframed boats or buoys, or discarded footage because of water conditions; this time, those elements remain in the edit.

The series features dolphins trapping fish by plugging holes with sea sponges, communication between fish species, and 200,000 crabs lining up to feed. White says real animals are probably more remarkable than AI could imagine.

Sources: The Verge

Read the original story ↗

TechCrunch compiled AI agents that work through text messages and can carry out tasks on a user’s behalf, including scheduling appointments, organizing calendars, researching trips, making reservations, sending emails and setting reminders. Some retain context and connect to apps and services people already use.

Examples include Ask Proxy, aimed at personal and work assistance; Caddy, publicly available in beta since April 2026 and integrated with iMessage and RCS; and Comma, which is open source and can be accessed through Signal, Telegram and WeChat. Fambot is designed to organize family life and works through an app, the web and SMS. The article also names Instinct as one of the most talked-about agents after a $1 billion funding round that, according to the article, brought its valuation to $10 billion.

Sources: TechCrunch

Read the original story ↗

Hugging Face and collaborators at the University of Pennsylvania developed LitPivot to support changes in research direction prompted by reading earlier work. The system separates a proposal into a problem, contribution and evaluation plan, organizes related papers around each part, and provides excerpt-supported critiques with revisions researchers can accept, edit or reject.

In a study with 17 researchers, domain experts rated ideas developed with LitPivot as better grounded in the literature than ideas produced with a text editor and an AI chat tool. Participants also viewed their revised ideas as more novel and useful and reported a stronger understanding of related literature. Hugging Face says LitPivot will be presented at UIST 2026 in Detroit.

Sources: Hugging Face

Read the original story ↗

Microsoft announced Building AI Skills and Agency Through Libraries, an initiative with the Public Library Association (PLA), a division of the American Library Association. The program will provide free AI literacy and skills resources and training to public library workers and patrons in the United States, focusing on practical skills, critical awareness and the safe and appropriate use of AI.

The initiative will combine Microsoft’s skilling resources with the PLA’s national network of expert practitioners. It will also offer opportunities for professionals to learn from one another through a Community of Practice, along with webinars and guidance. Microsoft says the training will cover how AI works, protecting personal and confidential information, critically evaluating AI-generated claims and sources, and risks related to safety and bias.

Sources: Microsoft

Read the original story ↗

Meta re-enabled Muse sharing after security researchers at Claroc reported that public links to conversations could be discovered through public DNS records. The company had paused the feature after being contacted about the issue and now says the flaw affected URLs for “artifacts” such as documents and images, not chat sharing itself. Previously shared artifacts received new URLs; anyone opening an old link will see a notice telling them to request the new address from the sender.

According to Wired, Claroc was able to enumerate 6,800 links and opened several hundred, finding information that apparently was not meant to be public, including an internal revenue review, a loan application and partial bank-account data. The researchers said the addresses were easy to discover because they included conversation titles in subdomains of cf.metaaiusercontent.com.

Sources: The Verge, Wired

Read the original story ↗

1Password launched 1Password for Agents, a tool that lets AI agents sign in to personal accounts without accessing passwords or other credentials. When an agent attempts to log in, a digital key verifies its identity, the user receives an alert and can approve the request; 1Password then fills in the credentials and completes the login on the agent’s behalf. The service supports agents including Anthropic’s Claude Duet, OpenAI’s Dots and ChatGPT Work, Grok Bot, Meta’s Muse and agents from Town, the company says, adding that it is working with other companies.

The launch comes as U.S. banks have begun detecting agent activity inside users’ accounts, according to Axios. Bankers interviewed by the outlet said they have difficulty distinguishing actions authorized by customers from unauthorized activity; a Chase spokesperson told Axios that customers should not give their usernames and passwords to anyone.

Sources: TechCrunch

Read the original story ↗

Uber and OpenAI announced an expansion of their partnership to bring ChatGPT Work to Uber’s entire global workforce of more than 30,000 people. According to TechCrunch, the company had tested the tool with 1,000 employees across areas including finance, operations, and product; about 80% of the early users were not engineers. They used it to analyze operational data, localize training materials, and build personalized interactive dashboards.

Uber has not yet settled on exactly how teams will use the tool. Sumanth Sukumar, the company’s vice president of engineering, told TechCrunch that Uber will give teams room to experiment and track the workflows they create and the resulting efficiency gains. Uber also said OpenAI worked closely with the company on the integration.

Sources: TechCrunch, OpenAI

Read the original story ↗

Microsoft says TK Elevator uses an agentic AI application, created with Microsoft Foundry and Azure OpenAI in Foundry Models, to guide technicians before maintenance visits. Through a phone or tablet, they can see which elevator to service, its floor, the reported issue, the tools to bring and the equipment’s service history. After the job, another voice agent asks questions, records the answers and requests more detail when needed, without requiring the technician to write or type.

The system uses a digital foundation TKE has developed over a decade by connecting elevators and escalators to a secure cloud and bringing together equipment data, service information and technical documentation in Microsoft Azure. Microsoft says TKE has 25,000 technicians caring for 1.4 million units worldwide; the company says its equipment carries 1.5 billion passengers a day. Technician Torben Teichmann said the AI works like a companion that gives advice at the job site.

Sources: Microsoft

Read the original story ↗

In an interview with MIT Technology Review, Devavrat Shah, Celonis’s chief scientist of enterprise AI, says AI agents need to understand operational context—including each task’s role, business constraints, and surrounding conditions—to support enterprise decisions. Without it, he says, systems may overlook important limits, make costly mistakes, or consume excessive computing resources. The Business Lab podcast episode was produced in partnership with Celonis.

Shah recommends connecting AI to the processes and data companies already use, with a contextual layer, guardrails, and human review. He suggests starting with a specific decision, testing its most difficult steps, and measuring gains in speed and value before scaling. According to Shah, automating repetitive tasks could free employees for strategic decisions, but organizations will need to invest in workforce education so people understand AI outputs and their implications.

Sources: MIT Technology Review

Read the original story ↗

Anthropic announced Claude Duet, an AI agent for professionals and small businesses in research preview. Available by waitlist to Pro and Max subscribers in the US, on iOS, Android, and the web, the system is built on Claude Opus 5.5, monitors connected apps, sends a morning brief, and performs tasks in a virtual machine with its own browser. It can create dashboards, financial forecasts, and web apps; according to TechCrunch, it connects to more than 6,000 connectors and plugins and can make online purchases, but it does not trade securities or move money between accounts.

The agent asks for approval before sensitive actions such as payments, and Anthropic says a new classifier will reject requests to share credit card numbers and government identification by default. Xataka notes that Duet arrives after Meta’s Muse and OpenAI’s Dots, focusing on work and productivity while competing agents are more focused on personal errands.

Sources: The Verge, TechCrunch, Xataka, Tecnoblog

Read the original story ↗

Walmart announced the Agent Gateway, a server based on the Model Context Protocol (MCP), to connect approved AI agents to its online inventory and let them make purchases on users’ behalf. The feature is available in beta, and Agent Profiles will arrive next month for Walmart accounts; a Walmart+ membership is not required. The company will also support shopping on Sam’s Club.

By default, each purchase will require approval, but customers can set a spending limit. The profiles will separate agent-initiated purchases from personal purchases and can be turned off or deleted at any time. Walmart says agents may not act beyond what the shopper authorizes, change the payment method or access more information than necessary. The company says its returns process also applies to purchases made incorrectly by an agent and that it will block systems that violate the rules. The initiative contrasts with Amazon, which began blocking shopping agents, including Perplexity and Meta’s Muse.

Sources: The Verge, Xataka, TechCrunch

Read the original story ↗

g1’s Fato ou Fake classified as false a video circulating on social media since at least October 15 that allegedly shows Leonel Brizola linking Jair Bolsonaro to militias and a death squad. The recording manipulates a segment from the first-round 1989 presidential debate hosted by SBT; according to the transcript reviewed by g1, Brizola does not mention Bolsonaro in the original material.

To trace the source, g1 used the InVID platform and a reverse search on Google Lens. Three tools detected manipulation: HiveModeration estimated a 99.4% probability that the audio had been altered with AI, while Hiya/Invid indicated a 99% chance that the audio was AI-generated, and Detector Mever/Invid indicated a 99% probability that the video was a deepfake.

Sources: g1

Read the original story ↗

Just Eat apologised after AI-generated images for restaurant food items were ridiculed on social media. The delivery platform had been testing the technology to illustrate dishes that did not yet have photos, but it produced inaccurate combinations, including bacon and eggs floating in a glass of fizzy drink, chips sticking out of a muffin and a cheese roll turned into sushi.

A spokesperson told the BBC that the company was removing the incorrect images and would stop producing new ones until the tool was further developed. Just Eat said it was trialling AI-enhanced images while working with tens of thousands of small restaurants. The incident also prompted criticism from customers who said they wanted to see what dishes actually look like to avoid disappointed expectations.

Sources: BBC

Read the original story ↗

Google has made conversational Gemini in Google Maps available to Apple CarPlay users in the US and says the global rollout is also underway. The assistant can handle multi-step questions, such as recommending places and then directing users there while avoiding toll roads, as well as helping find stops and report traffic disruptions. Drivers can activate it by tapping the Gemini icon on the car’s infotainment display.

The Verge notes that Google did not mention bringing to CarPlay the Personal Intelligence features announced for Gemini in Android Auto and cars with Google built-in, which can access information from apps including Gmail, Calendar, Maps, Photos, YouTube and Search. Google has not said whether that personalization will also come to CarPlay.

Sources: The Verge

Read the original story ↗

WikiProject AI Cleanup is a volunteer initiative created in 2023 to find and remove AI-generated material that violates Wikipedia’s rules. In an interview with IEEE Spectrum, founding member Ellia Lebleu said the signals, once marked by easily recognizable clichés, have become more structural: patterns such as the “rule of three,” negative parallelisms, and articles produced within minutes. Fake citations, or citations that do not support the claims they accompany, are also a major concern.

According to Lebleu, an analysis using the detection tool Pangram—described as a conservative lower-bound estimate—flagged AI in about 20% of new articles and drafts in late 2025 and 15% to 16% in early 2026. It is not yet clear whether the decline reflects Wikipedia’s new policies or more capable models. Volunteers are also using an AI-based source verifier, while the Wikimedia Foundation is developing Paste Check to identify text pasted directly from some chatbots.

Sources: IEEE Spectrum

Read the original story ↗

Lia Tax was created to help lawyers and accountants track tax rules, analyze documents, draft and review legal filings, and answer questions. Its launch is scheduled for October 29, according to MIT Technology Review Brasil. The founders say the platform uses a structured collection of official Brazilian tax data and underwent two months of blind testing, using an adaptation of a Stanford University methodology, to reduce hallucinations.

The startup says it fed the system more than 700,000 documents and plans to update the database daily in a field that receives about 100 regulatory acts a day. During roughly a year of development, the founders invested R$250,000 of their own money; there has been no outside investment yet. The team has five people, including CTO Moacir, and uses AI agents for financial and operational tasks.

Sources: MIT Technology Review Brasil

Read the original story ↗

Microsoft announced that Volkswagen Group equipped more than 330,000 employees with Microsoft Copilot. According to the company, the phased rollout began with Copilot Chat, which recorded more than 300,000 monthly active users; employees now also use Copilot with Work IQ across engineering, production, procurement and sales processes. More than 26,000 agents were active across the Group in the past month.

Microsoft and Volkswagen Group also intend to expand their collaboration and scale the Intelligent Data Foundation, the Group’s unified data ecosystem, on Microsoft Fabric. Microsoft says the initiative is intended to make trusted data easier to find, understand and use across the company.

Sources: Microsoft

Read the original story ↗