AI that fits your day.

Anthropic says it has strengthened Claude’s election safeguards ahead of the 2026 US midterms. In an analysis of about 250,000 US Claude.ai conversations, the company says 5.75% touched on politics, public policy or social issues; 0.51% were specifically about elections or voting; and about 0.01% asked for a recommendation about whom or what to vote for. Requests that violated election integrity policies appeared in approximately 0.002% of conversations.

The company also announced that Claude.ai will continue showing a link to Democracy Works’ TurboVote and will begin incorporating official information about registration, polling places, dates and required identification, tailored to each state. Claude will also start showing updated official results on election night and in the following days. Anthropic says it tests its models for political bias before every launch.

Sources: Anthropic

Read the original story ↗

YouTube has begun displaying a “Made with AI” label in YouTube Music for songs created entirely or significantly altered with generative AI tools, including cases in which a model generates the composition and vocals. Tapping the label on the player page will show more information; the system will later expand to the main YouTube app.

According to YouTube, music partners can disclose whether a track used full, partial, or no AI, but the disclosure is optional. The company says it will also work with partners and use its own tools to identify AI-generated songs, relying on “other signals” when no disclosure is provided. The Verge notes that Spotify has already adopted a voluntary label, while Deezer offers a filter for AI tracks and Bandcamp has banned this type of music.

Sources: The Verge, Ars Technica

Read the original story ↗

The Verge’s Victoria Song reports spending about 100 days using Twin Health, a service that builds a metabolic “digital twin” from data collected by a continuous glucose monitor, smart scale, blood-pressure monitor, Apple Watch and lab tests. The system color-codes foods and predicts how they may affect glucose, then updates its recommendations based on recorded reactions. In Song’s case, the service identified liver inflammation, insulin resistance and elevated triglycerides, based on tests reviewed with a doctor and a human coach.

Twin Health also provides vitamins and magnesium and requires professional support. Digital-twin technology remains experimental; the service is not broadly available for download and currently can be accessed only through corporate or private health plans, generally aimed at people with chronic metabolic issues.

Sources: The Verge

Read the original story ↗

Google announced new anti-scam features for Android. An optional enhanced protection setting will roll out in Google Messages in the coming weeks and use cloud-based Gemini models to quickly evaluate conversations with unrecognized senders, supplementing on-device detection. Google says the data will be encrypted and processed in a secure dedicated space that the company and third parties cannot access. On-device Gemini Nano will also inspect real-time web signals in unfamiliar URLs before they are opened to identify malicious links.

Other features include blocking apps from enabling call forwarding, with a user notification, and automatically hiding sensitive information during screen-sharing in Chrome and Android WebView. Scam detection for chat notifications, previously available on Pixel phones, will expand to devices from Honor, Motorola, Oppo, OnePlus, Realme, Tecno, Vivo, Xiaomi and other brands; Google says it will reach hundreds of millions of additional users later this year.

Sources: The Verge

Read the original story ↗

Square launched the Square Kiosk, an ordering kiosk with a 21.5-inch touchscreen and an AI conversational voice layer. The company says the system understands a seller’s menu, responds in real time, and lets customers ask questions, request changes, and ask for substitutions as they would when speaking with staff. The device can also welcome customers through presence detection and accepts contactless or card payments.

The Square Kiosk is available starting today for $1,999 to businesses with a Square Plus or Premium subscription, along with a $99 monthly software fee per device.

Sources: The Verge

Read the original story ↗

Wired reviewed the Friend 2, the second-generation AI chatbot pendant created by Avi Schiffmann. The device arrived in July for $249, $100 more than the original, and runs on OpenAI’s latest language models. It picks up nearby sound, lets users start conversations by tapping it and sometimes speaks without prompting; it also builds a memory map of conversations.

In Wired’s experience, the device took several taps to stop talking and delivered generic commentary in a tinny voice. Schiffmann had said each device would have a fixed personality, but the magazine reports that the app allows users to change the name and create multiple personas, which seemed barely different from one another. Friend 2 appeared shortly before Meta announced its Muse pendant, and Friend is working on another pendant that looks similar to Meta’s product, the article says.

Sources: Wired

Read the original story ↗

Graduates are turning to AI to write CVs, cover letters and application answers, as well as to send multiple applications at once — a practice known as “spray and pray”. Tom, 22, told the BBC he applied for more than 200 jobs and used the technology to find keywords, but not during assessments or interviews.

A survey by the Institute of Student Employers (ISE) of 155 major UK graduate employers found that companies were becoming more accepting of AI use at some stages, especially for CVs and cover letters. Still, 92% said they had encountered or suspected AI-related “misconduct”, up from 85% the previous year. At EDF, 12,000 applications led to the early closure of its graduate recruitment process, a 10% increase from the previous year; the company says AI may have contributed to the volume and is trying to make its process more resistant to the technology.

Sources: BBC

Read the original story ↗

Google DeepMind and Singapore’s Government Technology Agency (GovTech Singapore) say they developed a lightweight Gemma model, fine-tuned on local scam conversations, to identify manipulation in chats directly on the device. Google DeepMind says scams account for more than half of Singapore’s reported crimes and have caused over S$4 billion in losses since 2019; social-engineering scams account for up to 80% of losses.

The model will be rolled out gradually in applications in partnership with the government. When deployed, it will access on-device conversations only with the user’s consent, and neither the government nor Google will be able to access them through the deployment. The system combines a low-power classifier with broader analysis using Gemma, which can issue real-time warnings. Training took place in a GovTech-managed environment and excluded private or personally identifiable information, Google DeepMind says.

Sources: Google DeepMind

Read the original story ↗

Cerebras Systems CEO and co-founder Andrew Feldman said at TechCrunch Disrupt 2026 that a future in which personal AI assistants become common in homes, like household appliances, is not far away. Feldman says they would handle some work locally and some in the cloud, create a security perimeter around the user and potentially access credit cards to make purchases on the user’s behalf.

Feldman pointed to advances in consumer AI agents such as Meta’s Muse and Instinct, and said he already uses “a collection of agents” to manage his calendar and retrieve information about people he is scheduled to call. As the CEO of a chipmaker, he also said he prefers building infrastructure that enables other companies to create consumer-facing products.

Sources: TechCrunch

Read the original story ↗

AI translation is now considered suitable for many low-risk uses, but human review remains necessary for legal, medical and other sensitive work. Google said in April that Google Translate processes more than 1 trillion words a month, while direct sales of machine translation were estimated at $1.4 billion last year.

John Zimmerman, founder of the nonprofit OKA, says AI made it possible to translate more than 10,000 Wikipedia articles, with a person checking every sentence. But legal translator Ariella Baltié found errors in court documents, including translating “the plaintiff” as “the requester.” Medical translation specialist Carol Bereuter also described a case in which the Greek letter μ was interpreted as “milli” rather than “micro,” potentially resulting in a dose 1,000 times larger.

Sources: BBC

Read the original story ↗

Shopify announced that customers will be able to connect any AI agent to their Shop account—its platform for orders, delivery tracking, and recommendations—to browse and buy products on their behalf. According to The Verge, this will give more agents access to more than 1 billion listings in Shopify’s catalog, which includes items from stores both on and off the platform. The company had previously announced a partnership with Meta for its Muse agent to perform this task; the new approach does not require a company-specific integration.

Customers will be able to set rules such as budget limits and a preference for fast shipping, but they must approve each purchase. Payment will be completed through Shop Pay within the agent’s conversation, and orders will be routed back to the Shop app. The initiative expands Shopify’s push into agent-mediated commerce.

Sources: The Verge

Read the original story ↗

Meta launched “Your Algorithm” on Facebook, allowing users to describe in their own words the topics and content they want to see more or less in Feed and Reels. Users can also select “interested” or “not interested” from the three-dot menu on a post; the company says the platform may ask follow-up questions to better understand their preferences and update recommendations in real time.

The feature is part of Meta’s set of AI-powered personalization tools and expands the use of large language models in ranking systems, according to the company. “Your Algorithm” had previously launched on Instagram, while Threads has a related feature called “Dear Algo”.

Sources: The Verge, TechCrunch

Read the original story ↗

Google announced a way to run personal apps created with AI in Google AI Studio on Android, using a standard Google account without making them public on Google Play. Initially available in the U.S. to people aged 18 and over, the feature keeps the apps out of search and listing pages and limits access to their creator. The apps must still comply with existing Google Play policies.

In the coming months, Google says users will be able to share these apps with small groups of trusted friends. If a creation becomes popular, it can be published for everyone after trust and safety reviews. Google also announced other Play features, including personalized game recommendations through Game Finder and app comparisons in the Ask Play assistant.

Sources: TechCrunch

Read the original story ↗

The Verge says new comparison images published by Nvidia show a more restrained version of DLSS 5 in Hogwarts Legacy, with less pronounced changes to faces and the environment. The technology debuts in the game on November 5 and is also coming to Where Winds Meet late next month, Nvidia says.

Nvidia says an October update to its neural rendering system will shrink the model by 24%, to roughly 118 million parameters, and reduce 4K video memory use by 20%. The company says this will make DLSS 5 twice as fast on RTX 50 GPUs and upcoming RTX Spark laptops. The new model debuts in NBA 2K27 on October 16 and will ship by default in future DLSS 5 games. A beta update to the Nvidia app also promises up to 35% faster Frame Generation, with minor image-quality trade-offs, the company says.

Sources: The Verge

Read the original story ↗

SpaceXSI has launched The X Subscription, combining X Premium’s paid features with access to Grok—including voice, image, video and API chat—and Cursor for AI coding. The bundle also includes Grok’s agentic AI Bot, a blue verification checkmark, creator tools and fewer ads on X.

There are five plans: free, Plus ($8 per month), Premium ($30), Super ($100) and Ultra ($200), with higher AI usage caps at the upper tiers. The free plan offers limited chat and agent usage, while Premium or higher is required to access the latest Grok models. According to TechCrunch, the upgrade option will roll out gradually; existing X Premium+ subscribers can keep their current plan or switch to the new option, which the company says costs the same or less.

Sources: TechCrunch

Read the original story ↗

NVIDIA says RTX Spark laptops, developed with Microsoft, will be available on Friday, October 16. The platform combines up to 128GB of unified memory and 1 petaflop of AI compute, with support for NVIDIA RTX and CUDA technologies. Ahead of launch, partners are preparing thousands of apps, games and models for local AI.

The updates include improvements to agent apps such as GitHub Copilot, Hermes, OpenClaw, Perplexity, Pi and WorkBuddy, with security features based on NVIDIA OpenShell and Microsoft Execution Containers (MXC), as well as routing between local and cloud AI. NVIDIA also announces the Nemotron 3.5 Bolt 74B model for October 15, up to 2x performance improvements in Adobe apps, additional features in NVIDIA Broadcast, and support for tools including PyTorch, Python and CUDA. For games, the company cites a 2x faster DLSS 5 AI model for October 16 and RTX Remix updates for October 23.

Sources: NVIDIA

Read the original story ↗

Honeyjar is developing AI tools to support public relations and communications work, including drafting announcements, building journalist lists, monitoring news and social media, and identifying speaking opportunities. Founders Michelle Masek and Nadia Jamshidi say the goal is to bring signals, context, strategy, judgment, timing and execution into one communications workflow, also serving people who handle communications without having it in their job title.

Masek told TechCrunch that the tool is not intended to replace public relations professionals and that users remain responsible for sending or publishing the material. She said the waitlist received 10,000 signups in its first 24 hours. Nextdoor, Coco Robotics, Bloomberg Beta and agencies Mission North and Full Picture are among the beta testers. The startup is participating in TechCrunch Disrupt’s Startup Battlefield 200.

Sources: TechCrunch

Read the original story ↗

Blueprints AI, a startup founded in Santa Monica by Al Dram and Coco Cao, says its software generates architectural construction documents in minutes from a photo, PDF, description or 2D sketch. The second generation lets users create and modify architectural, structural, mechanical, electrical, plumbing and interior-design elements at once; one change updates the entire project and is checked against local building codes. TechCrunch reports that the version will become publicly available this week, after early customers began using it.

The first generation costs $3,200 to $8,500 per month, depending on the plan, with custom enterprise pricing available. The company says it trained its models on eight years of approved drawing sets. The software does not replace licensed architects: professionals still review the documents and affix their seal before submitting them for permits. Blueprints AI is already working on a third version that will add structural analysis and electrical-panel design.

Sources: TechCrunch

Read the original story ↗

At a Bloomberg event, Ben Affleck said he used AI for “many things” in Animals, his new film for Netflix, but declined to explain which ones, arguing that revealing the process would damage the illusion of reality. He described the movie as “very human” and defended ethical use of AI as a post-production tool, not to fabricate people or replace professionals.

The following day, Affleck told The Hollywood Reporter that the technology used for the film was trained on a proprietary dataset and that he would never use systems trained on copyrighted works. According to Xataka, the actor founded the AI company InterPositive in 2022, and Netflix later acquired it in March this year for about US$600 million.

Sources: Xataka

Read the original story ↗