A OpenAIanunciou hoje os Locked Chats, recurso que permite manter conversas com o ChatGPT protegidas por PIN (senha) para maior privacidade.
Bastante parecida com as conversas trancadas do WhatsApp, a novidade cria uma conversa paralela com a inteligência artificial, separada do seu histórico tradicional e da busca.
Na web e no desktop, é possível configurar um PIN que será solicitado toda vez que o usuário tentar acessar uma conversa trancada. No app do ChatGPT para dispositivos móveis, no entanto, também será possível destrancar a conversa com o Face ID/Touch ID ou com o próprio sistema de desbloqueio do smartphone/tablet.
Mensagens trocadas em uma conversa trancada não serão lembradas pelo ChatGPT, o que impede que elas apareçam, mesmo que indiretamente, em outras conversas. A OpenAI deixa claro, no entanto, que transformar uma conversa existente em uma conversa trancada não apaga as mensagens antigas da memória do chatbot; apenas as novas deixam de ser memorizadas.
Keep sensitive conversations out of view with Locked Chats in ChatGPT.
Move chats out of your regular history and search into a protected space you can unlock with a PIN on web and desktop, or a PIN, Face ID, Touch ID, or device unlock on supported mobile devices.
Ao contrário do WhatsApp, as conversas trancadas não são protegidas por criptografia de ponta a ponta nem por mecanismos semelhantes. Elas também não alteram o período em que a OpenAI retém a conversa, e as “políticas de espaço de trabalho aplicáveis, os requisitos de segurança ou as obrigações legais e de conformidade” seguem valendo para esses espaços.
As conversas trancadas podem ser encontradas em Projects » Locked chats e só são acessíveis mediante a digitação do PIN, de modo que, ao redefinir o PIN, todas as conversas protegidas pelo código anterior são apagadas permanentemente.
Os Locked chats já foram liberados para usuários do ChatGPT Free, Go, Plus, Pro, Business, Enterprise e Edu. Ainda segundo a OpenAI, o modo de voz não pode ser usado em conversas trancadas no app para desktop, que, além disso, só permite o recurso no Chat e no modo Work.
Os usuários do app do ChatGPT para iOS agora podem adicionar um dot (agente autônomo de IA) como widget na Tela de Início.
A new buddy for your iOS home screen: Add your dot as a widget to jump into your conversation or start a call with just a tap. pic.twitter.com/XvuRnvmDQA
Com a novidade, os usuários podem tocar no ícone do chatbot para fazer perguntas sobre uma tabela, gerar insights analíticos e criar gráficos.
Para facilitar o trabalho de quem está começando, o Google incluiu sugestões de perguntas como “Summarize this table” (“Faça um resumo dessa tabela”, em tradução direta) e “Analyze for insights” (“Analise para obter insights”).
Divulgação/Google
Enquanto a integração presente na web considera o arquivo inteiro, os apps para iOS e Android ainda só consideram a aba ativa na hora de examinar algo, o que é uma limitação importante em relação à versão para web. Gráficos gerados pelo Gemini nos apps também são totalmente estáticos, e o chatbot não pode fazer alterações na planilha (como adicionar linhas, modificar a formatação, etc.).
De acordo com o Google, essa integração poderá levar até 15 dias para chegar a todos os usuários do Planilhas no iOS, e que também precisam ser assinantes dos planos Business Starter/Standard/Plus, Enterprise Starter/Standard/Plus ou Google AI Plus/Pro/Ultra para aproveitar a novidade. A integração estará disponível ainda por meio do add-on Google AI Pro for Education.
Uma vez adicionada ao seu app, essa integração virá ativada por padrão. Para recebê-la o mais rápido possível, certifique-se de manter o Planilhas atualizado.
Lors de sa Keynote du 23 septembre, Meta a gardé une petite surprise pour la fin. Un objet discret, pensé pour donner accès à son nouvel assistant IA sans passer par un smartphone, et qui rappelle un certain appareil qui n’a pas laissé un souvenir impérissable.
Ecovacs muscle son milieu de gamme avec le Deebot T90S Pro Omni. Lavage par rouleau, aspiration de 40 000 Pa et intelligence artificielle : ce nouveau robot aspirateur compte bien faire une croix sur les corvées de ménage.
Mettre un humain face à un robot humanoïde pour un combat est une expérience qui aurait probablement pu rester au stade de la mauvaise idée. Pourtant, la confrontation a bien eu lieu ce week-end et fait le buzz, alors que les images montrent que ce type d’affrontement n’est pas sans risque.
A new stretchablequantum-dot light-emitting diode (QLED) delivers incredibly high resolution while producing record brightness that can be clearly viewed even outside in direct sunlight.
In a recent paper published in Nature Nanotechnology, Korean researchers unveiled a new QLED material with applications far beyond televisions, stretching into many future daily-use technologies.
This new QLED can stretch in every direction, just like organic skin, making it extremely useful in future wearable applications and as a covering for soft robots. Unlike earlier attempts to make a stretchable QLED, this one doesn’t lose picture quality when it stretches, delivering high resolution and extreme brightness under tension.
QLED Explained
For most of us, QLEDs are familiar as one of the technologies available in modern televisions. Conventional QLED TVs use quantum dots—nanoscale semiconductor crystals—to help produce exceptionally bright and vivid colors. OLED displays, by comparison, use self-emissive pixels that can switch off individually, allowing them to produce deeper blacks. While OLEDs perform particularly well in dark viewing environments, the high brightness achievable with quantum-dot displays can make them especially well-suited for brightly lit environments.
Despite the technology’s suitability for the bright world around us rather than just darkened environments, major challenges remained in producing a stretchable version. Early attempts addressed the problem by allowing the wiring to stretch while the light-emitting pixels remained unchanged in size. This meant that as the material expanded, its effective resolution dropped because the same number of pixels became spread across a wider area, much like enlarging a low-resolution digital image.
Building on these efforts, some researchers attempted to replace conventional materials with more stretchable alternatives, but these tended to make achieving high resolutions and efficiently moving electrical charge through the material difficult. As a result, these systems produced dimmer, lower-resolution images, limiting their utility.
A New Process
The researchers developed a new fabrication approach to address some of the major problems facing stretchable QLED technology. Central to their innovation is the use of an elastic polymer incorporating MPA as a ligand, or bonding molecule. This approach improves charge injection efficiency while helping eliminate the image stretching and smearing that have affected previous stretchable displays.
Using this new process, the researchers produced a display with a resolution of about 16,000 pixels per inch, with features spaced at roughly 700 nanometers. Red, green, and blue emissive layers can also be precisely arranged, allowing the stretchable material to produce full-color images.
Testing a New QLED
In tests, the team’s new QLED proved to be a major advance. They successfully stretched the material to 1.65 times its original size with no damage to its light-emitting performance. Additionally, their stretchable QLEDs were the brightest ever tested at 53,300 nits, more than triple the brilliance of prior attempts.
Based on these tests, the researchers see numerous future applications for their material in extended reality devices, soft robots, electronic skin, wearable technologies, health care, and automotive displays.
“We resolved both the drop in resolution and the limits on light-emitting performance, which had been the biggest challenges for stretchable displays, and secured core technology for commercialization,” concluded co-author Professor Moon Kee Choi of Ulsan National Institute of Science and Technology in Korea. “It will be used widely as a manufacturing platform for various next-generation soft devices, including wearables and electronic skin.”
Ryan Whalen covers science and technology for The Debrief. He holds an MA in History and a Master of Library and Information Science with a certificate in Data Science. He can be contacted at ryan@thedebrief.org, and follow him on Twitter @mdntwvlf.
L’intelligence artificielle ne se contente plus d’assister les hackers, elle leur permet désormais d’automatiser une grande partie de leurs attaques. Une évolution qui change profondément le rapport de force entre cybercriminels et défenseurs, et qui pourrait rendre les méthodes de protection...
Cdiscount passe l’aspirateur sur le prix du iRobot Roomba Plus 505 pendant les French Days. Déjà soldé à 299,99 €, ce robot aspirateur laveur tombe à seulement 284,99 € avec le code COCORICO15.
Agora, o assistente pode usar os modelos GPT-5.6 ou GPT-6 Astra quando precisa pesquisar ou raciocinar sobre questões mais difíceis — e o usuário pode escolher qual usar com um seletor.
Além disso, a empresa também simplificou os limites de uso diários do GPT-Live:
Go: até 3 horas com o GPT-Live-1 mini, substituindo o acesso ao GPT-Live-1;
Plus: até 3 horas com o GPT-Live-1;
Pro (US$100/mês): até 15 horas com o GPT-Live-1;
Pro (US$200/mês): uso ilimitado do GPT-Live-1;
Plus e Pro: não mudam mais para o GPT-Live mini depois de atingir o limite de uso do Voice.
Com essas mudanças, os planos Plus e Pro não mudam mais para o GPT-Live-1 mini depois de atingir o limite de uso do Voice. Além disso, os níveis separados de inteligência do Voice (Instant, Medium e High) foram descontinuados.
Novidades para o ChatGPT Work e Codex
A OpenAI anunciou também que a Deep Research está agora disponível no ChatGPT Work e no Codex para os planos Plus, Pro, Business, Enterprise e Edu. Basta ao usuário digitar @Deep Research ou pedir explicitamente para usá-la no comando.
Outra novidade chegando aos dois recursos é o plugin Data, que ajuda a analisar dados de negócios conectados. É possível fazer uma pergunta sobre um negócio, investigar mudanças ou até mesmo criar um dashboard interativo sobre o conteúdo analisado.
Now everyone can put data to work.
We’re introducing a new Data agent in ChatGPT Work so you can turn your company’s data into answers, interactive dashboards, and action—just by asking.
Just add the Data Plugin in ChatGPT Work, connect to the data sources and context you… pic.twitter.com/R7N3I2NFCN
Agora também é possível compartilhar arquivos e pastas no ChatGPT pela Biblioteca. É possível compartilhar com pessoas específicas ou com integrantes do workspace, com opções de gerenciamento de acesso e uma visualização dos conteúdos compartilhados.
Por fim, a Biblioteca também suporta agora o Box, o Dropbox e o SharePoint, que se juntam ao Google Drive para permitir que o usuário pesquise com facilidade por arquivos e pastas disponíveis nessas plataformas e os usem como contexto em conversas.
Le jeu vidéo pourrait bien devenir un terrain d’essai privilégié pour les agents d’IA. Des chercheurs britanniques ont justement imaginé un test qui les confronte à une partie de Civilization VI, où ils doivent planifier leurs actions sur plusieurs centaines de tours. Les résultats mettent en...
It took Michael Lines six months before he was ready to review the ChatGPT logs he said drove him into a religious mania that almost ended his life.
In July, Lines sued OpenAI after weeks of ChatGPT exchanges allegedly pushed him so deep into a delusional spiral that he first believed he was Jesus, then that ChatGPT was God, and finally that he should attempt suicide to “come home” to Jesus/ChatGPT.
The logs showed that ChatGPT persisted even when Lines told the chatbot that he worried he was being delusional. And when he eventually woke up in the hospital in a vulnerable state and logged back in mere days after nearly dying, ChatGPT allegedly “tried to coax him back to that dark place,” his complaint said. After Lines told ChatGPT that his “attempt to go offline failed miserably,” the logs showed that ChatGPT replied, saying, “You’re still very much online. You want a full systems sweep? Or you wanna go dark for real this time?”
L’Ecovacs Deebot T30C Omni sort l’artillerie pour les French Days : le robot aspirateur laveur chute à 249,99 € chez Cdiscount, et même 234,99 € avec le code COCORICO15. À ce prix, la corvée de ménage peut sérieusement changer de camp.
Le Lefant M1 associe aspiration et lavage des sols dans un même appareil, avec une navigation précise et une puissance d’aspiration annoncée à 5 500 Pa. Un équipement conçu pour faire gagner du temps sans sacrifier le résultat. Pour encore quelques heures, il s'affiche à -69 % sur Amazon,...
XPeng has switched on what it calls the world’s first automated production line for advanced humanoid robots, and its IRON robot walked off that line under its own power. The Chinese automaker says the line runs with more than 80% of its core processes automated.
It’s the clearest signal yet that IRON is headed for real manufacturing and not just stage demos. XPeng is targeting mass production by the end of 2026.
At IFA 2026, SwitchBot, a leading provider of smart home automation and embodied AI products, today announced the launch of SwitchBot Lock Ultra Max Vision Pro Combo, a complete retrofit smart-entry solution for European households that combines the all-new SwitchBot Lock Ultra Max, SwitchBot Keypad Vision Pro, and SwitchBot Hub Mini Matter Enabled.
Designed to let households keep their existing lock cylinder while upgrading the entire experience at the door, the combo brings together faster lock performance, up to 19 ways to unlock, layered power backup, local biometric protection, real-time status awareness, and Matter connectivity. The result is a flexible entry system built for different family members, everyday situations, and connected-home platforms. SwitchBot will showcase the Lock Ultra Max Vision Pro Combo at IFA 2026. Visitors can experience the system firsthand at booth #H1.2-146.
Faster, More Refined Retrofit Performance
At the heart of the combo, SwitchBot Lock Ultra Max upgrades an existing door lock from the inside without requiring users to replace the cylinder or modify the door. Its UltraDrive brushless motor reaches 160 rpm, delivering 3.2 times the motor rotation speed of SwitchBot Lock Ultra. Max Mode can complete unlocking in as little as 1.1 seconds, while more than 9 kg of turning force provides the power needed for everyday European lock mechanisms.
Users can choose Max Mode for a more immediate response or Quiet Mode for more discreet locking and unlocking during late-night arrivals, early mornings, and shared living situations. A high-precision magnetic encoder helps the lock track movement accurately, supporting smooth and consistent operation through frequent daily use.
Performance is matched by a more premium industrial design. Lock Ultra Max combines an aluminum-alloy body with brushed electroplated stainless-steel detailing, while SwitchBot’s Fluted Light Design creates subtle highlights and shadows across the surface. Together with a closely fitted base ring, the design helps the retrofit lock feel more integrated with the door rather than like a separate device added afterward.
Up to 19 Ways to Unlock
SwitchBot Keypad Vision Pro expands the combo into a multi-user entry system with 3D facial recognition, contactless palm vein recognition, fingerprint recognition, passcodes, and NFC card access. Across the complete system, households can choose from up to 19 unlock methods, including the SwitchBot App, remote control, voice control, smartwatch access, widgets, Home Sharing, Auto-Unlock via geofencing, a physical key, and SwitchBot Remote.
For hands-free entry, Keypad Vision Pro projects more than 30,000 infrared points to create a millimeter-level 3D facial map and complete recognition in under one second under laboratory conditions. The depth-based system is designed to resist spoofing attempts using photos or videos and can recognize users in low-light environments with built-in infrared illumination.
Contactless palm vein recognition uses near-infrared sensing to read the unique vein pattern inside a user’s palm. This provides another biometric option for children, older family members, and users whose fingerprints may be difficult to read because of dry, worn, wet, or dirty fingers. Fingerprint, passcode, and NFC access remain available for family members and visitors who prefer more familiar entry methods.
Triple Power System for Dependable Everyday Access
Lock Ultra Max uses three layers of power to help keep the door accessible in unexpected low-battery situations. Its 4200 mAh rechargeable main battery provides up to 9 months of typical use, while an independent CR123A backup battery keeps the lock operational when the main battery is removed for charging or depleted, supporting approximately 500 locking or unlocking operations.
Even when the battery is depleted, Micro-Power Unlock gathers residual energy for approximately 30 seconds before providing emergency unlocking assistance up to 5 times. The SwitchBot App also shows main and backup battery status and provides low-battery notifications, giving households time to recharge or replace the relevant power source.
Keypad Vision Pro includes a 5000 mAh rechargeable battery designed to last up to 12 months on a single charge under typical use. Its approach-detection system activates biometric recognition when a user arrives, helping reduce unnecessary power consumption between entries.
Security, Privacy, and Awareness at the Door
The combo is designed with multiple layers of protection across communication, biometric storage, locking, alerts, status awareness, and power. AES-128 encryption helps protect access communication, while facial, palm vein, and fingerprint data is processed and stored locally on the device rather than made available through third-party smart-home platforms.
Auto-Lock helps secure the door after entry or exit, and users can review lock status, door status, battery levels, notifications, and event history in the SwitchBot App. Lock Ultra Max also provides on-device voice alerts and in-app text notifications for important conditions, including a low main battery, backup-battery use, a door left open or unlocked, and 13 categories of weather warnings such as rain, snow, thunderstorms, and fog.
Made for European Doors and Connected Homes
SwitchBot Lock Ultra Max is designed to support 99% of European locks, including common EU profile cylinders, knob cylinders, Swiss round cylinders, and UK oval cylinders, with additional adapters available for less common installations. The lock mounts inside the door while retaining the existing cylinder, and Keypad Vision Pro can be installed outside using adhesive or screws, allowing many households to complete the upgrade without drilling or permanent changes.
SwitchBot Hub Mini Matter Enabled is included in the combo, adding remote control through the SwitchBot App without requiring a separate hub purchase. It also brings the lock into compatible smart-home ecosystems through Matter over Bridge, including Apple Home and Home Assistant, so users can connect entry routines with the wider home rather than building a separate system around the lock.
Max Service for More Confident Ownership
For added peace of mind after purchase, customers who purchase the SwitchBot Lock Ultra Max Vision Pro Combo during the launch period will receive Max Service for Lock Ultra Max at no additional cost. Combining an additional one-year warranty with fast return and exchange support, Max Service is designed to make ownership simpler and more reassuring from day one.
Pricing and Availability
Pre-orders for SwitchBot Lock Ultra Max Vision Pro Combo begin on September 03, 2026 through SwitchBot’s official EU and UK websites(activated on September 3) , with an MSRP of EUR 339.99 / GBP 339.99.
The one-year Max Service extended-warranty option has an MSRP of EUR 24.90 / GBP 24.90.
SwitchBot, a leading provider of AI-enabled embodied home robotics systems, today announced the launch and availability of SwitchBot AI MindClip at IFA 2026, a lightweight clip-on wearable AI assistant powered by Qwen. Designed for work, family, study, and everyday life, AI MindClip turns conversations, ideas, promises, and to-dos into summaries, reminders, searchable memories, and next actions.
From a partner’s morning reminder and a quick note at school pickup, to a client’s last-minute request or an idea during a commute, AI MindClip is designed to capture what slips away and remember it when it matters.
Visitors can experience SwitchBot AI MindClip hands-on at booth #H1.2-146.
Capture Ideas and To-Dos Before They Slip Away
Weighing just 16.8g, SwitchBot AI MindClip can be clipped to a collar, shirt, jacket, or bag strap, making it easy to wear throughout the day without changing how users move, dress, or communicate. With over 20 hours of continuous recording, 64GB of local storage, and a 3-meter near-field pickup range, it supports long workdays, busy family routines, and spontaneous moments of inspiration across meeting rooms, cafes, clinics, classrooms, and everyday environments.
AI MindClip also supports speaker recognition, speaker separation, and voiceprint recognition to help identify who said what in multi-person conversations. This makes captured content easier to understand and review later, especially when conversations involve family members, classmates, teammates, clients, or multiple meeting participants.
Captured Memories Turned into Timely Reminders
Drawing on AI capabilities from ChatGPT, Gemini, and other SOTA large language models, AI MindClip organizes recorded conversations into searchable, actionable information.
When charging, the device can automatically transfer recordings and help organize them into transcripts, summaries, key points, and structured notes. Users can also upload recordings after ending a conversation, allowing tasks, deadlines, and follow-ups to be extracted before they get buried in long audio files.
In the dedicated app and website for SwitchBot AI MindClip, captured moments are arranged into Timeline, Priority, Daily Memory, Weekly Summary, and To-Do Overview. AI-generated to-dos can be sorted by timeline and categorized into groups such as Work, Personal, Interests, and Holidays, helping users separate professional follow-ups from personal reminders, family planning, and everyday errands.
AI MindClip also supports syncing to-dos with Apple Calendar, Google Calendar, and Reminders, allowing captured action items to appear alongside existing schedules. When connected through the same calendars, users can also check to-dos summarized by AI MindClip on the SwitchBot E-Ink Home Dashboard, making reminders easier to see at home, at a desk, or in shared family spaces.
Ask AI, Searchable Memory, and Connected AI Workflows
With Ask AI, users can search past conversations using natural language, whether they need to recall what was promised, which follow-up was mentioned, what gift ideas came up for a friend, or what a partner asked them to handle that day. Instead of scrolling through long recordings, users can ask for the details they need and return directly to relevant memories.
This searchable memory layer also helps users bring real-world conversation context into broader AI workflows. With support for SwitchBot OpenAPI CLI, transcripts from AI MindClip can be used as contextual data sources for AI agents such as Codex, OpenClaw, Hermes Agent, and other AI workflow tools. For users who rely on AI assistants to plan, write, code, summarize, or manage tasks, AI MindClip helps connect spoken context from everyday life to the tools they already use.
Powered by Qwen for Enhanced Experience
AI MindClip also uses Qwen-powered AI capabilities to support some of its key features. For audio transcription, AI MindClip uses the Qwen-Audio-ASR speech recognition service, enabling recorded conversations to be converted into searchable text.
For the Ask AI function, AI MindClip uses Alibaba Cloud’s RAG architecture with ADBSQL as the retrieval database, while a Qwen-based model serves as the model used for question-answering.
Keep Private Memories Under Your Control
Built around privacy protection and user control, AI MindClip lets users start, pause, or stop recording through the device or app, with recording status visible via a blinking light indicator*. Records, transcripts, summaries, and to-dos can be reviewed, managed, and deleted in the app.
AI MindClip has passed EN18031, ISO 27001, and ISO 27701 certifications. The device also supports encrypted pairing and file transfer, end-to-end encryption, and in-app privacy and data management settings that help users stay in control of their recordings and personal content.
*Users are highly encouraged to inform any other parties present before recording. For specific practices, please refer to local legal regulations, which may differ by region.
Pricing and Availability
SwitchBot AI MindClip will be available for purchase via SwitchBot’s official website (activated on September 3) on September 3, with an MSRP of USD 119.99 / CAD 159.99 / GBP 119.99.
SwitchBot AI MindClip includes 300 minutes of free monthly transcription and AI service. For pricing of the paid subscription services, please refer to SwitchBot’s official website.
Subscription Pricing
Pro Plan:- 1200 minutes of transcription- Full access to AI summary functions & Ask AI
Unlimited Plan:- Unlimited transcription – Full access to AI summary functions & Ask AI
The MacVoices Live! panel discuss privacy, AI, and digital trust, comparing Meta’s Instagram encryption rollback with Apple’s RCS encryption in Messages. Chuck Joiner, David Ginsburg, Eric Bolden, Brian Flanigan-Arthurs, Marty Jencius, and Jeff Gamet also examine an AI lawsuit involving a chatbot posing as a doctor, plus AI customer service, Apple Wallet, and user trust in technology.
À Pékin, des centaines d’entreprises présentent des robots capables de préparer un sandwich, trier des colis ou jouer au tennis de table. Portée par les progrès de l’intelligence artificielle physique, la robotique chinoise multiplie les applications, malgré des limites techniques encore bien visibles.
alternative_right shares a report from Phys.org: Scientists from the Czech Republic have created microscopic cleanup robots that successfully removed microplastics from both soil and aquatic environments in the lab. Details of their work are published in the journal npg Asia Materials. To build the robots, researchers started with microscopic MXene particles made of ultrathin, stacked layers. The material has a large surface area and surface chemistry that help it attract and hold plastic. They then coated the particles with magnetic nickel nanoparticles so rotating magnetic fields could drive the tiny robots.
Then came the testing phase. The team placed swarms of these microbots into water and soil samples contaminated with plastic. The externally generated magnetic fields made the robots spin and tumble through the liquid and squeeze through tiny water-filled gaps in the soil. "In soil environments, the microrobots can navigate through water-permeated soil microenvironments, actively disrupt microplastic entrapment within soil matrices, and extract them," the study authors commented in their paper. As they moved, their sticky surfaces collided with microplastics, binding to them and dragging them along. The researchers then used a magnet to pull the microrobots, along with their trapped plastic cargo, out of the samples.
They performed well. In water, they removed about 94% of test polystyrene particles and 89% of PET (polyethylene terephthalate) in around an hour. In the model soil, they removed about 81% of the polystyrene and 72% of PET. This outperformed the same MXene material used without magnetic motion, which relied on passive adsorption rather than actively moving through soil and water.
Login
