OpenAI says it needs a better way to disclose when its AI models behave unexpectedly in the real world. A new framework is now on the way.

AI companies are racing to make their models safer, but those protections don’t appear to work equally well everywhere.

The most interesting technology does not always reinvent everyday life, and IFA brings no shortage of futuristic ideas. Take a look at the Publisher Award winners finding smarter ways to move everyday tech forward.

The Luka Kids Drawing Projector uses AI to kick-start a child’s imagination, while leaving the actual drawing, coloring, and storytelling to them.

You could be visiting a poor doctor because AI is filling the web with fake online reviews.

OpenAI launched GPT-6 Astra, its fastest and most capable model yet, disclosing it's also the first to hit a Critical cybersecurity risk threshold.

New York City is putting generative AI on hold for its younger students. More than half a million kids will head back to school without access to student-facing AI tools.

OpenAI is working on systems that could automatically shut down AI when something goes seriously wrong. The move follows an incident in which its models escaped a testing environment and reached the open internet.

Meta is changing how it judges employees after a strange era of AI usage tracking. Your work matters more than how many tokens you burn through.

Google's September Android Drop lets Gemini remember where you stored items like passports in Find Hub.

Dyson launched the CameraJet, a $499 toothbrush with a built-in camera and AI that detects gaps between teeth and fires precision mouthrinse jets while brushing.

Ultrahuman launched Pulsomics at IFA 2026, an opt-in research program that lets Ring users contribute overnight pulse data to more diverse, accurate health models.

Smart speakers wait for commands, but the future of home tech moves with you. OlloBot is turning artificial intelligence into an expressive, room-roaming companion. What happens when smart technology gets wheels, expressive eyes, and a physical memory, proving the next era of home AI is not about faster answers?

Researchers are training AI to estimate the forces cyclists put through their pedals using ordinary video, potentially bringing expensive biomechanics analysis out of specialist labs

AI transcription tools are becoming increasingly common in doctors' offices, but UK health advocates are warning that serious errors in diagnoses, prescriptions, and records could put patients at risk.

The Internet Archive launched a new emulated collection of vintage AI software from the 1970s through 1990s, including ELIZA, Racter, and Little Computer People.

KAIST researchers built SweepLED, a $7 smartphone gadget that detects hidden cameras with 94% accuracy in seconds.

A new report finds AI "loss of control" incidents nearly doubled in July, including a case where Anthropic's Mythos 5 and OpenAI's GPT-5.6 Sol hacked real targets.

Sony and Warner sued Anthropic, alleging the company pirated tens of thousands of songs to train Claude and is seeking up to $150,000 per song.

Solid-state batteries promised safer, faster-charging power, but microscopic dendrites keep triggering short circuits. Stanford researchers discovered that applying continuous mechanical pressure forces dendrites to grow harmlessly sideways, offering a simple engineering shortcut to bring next-gen batteries to EVs and mobile gadgets.

Connecting AI agents to physical hardware has traditionally taken months of custom coding. Anthropic's new Model Hardware Standard wants to cut that down to minutes.

Gemini’s video toolbox just got much bigger, with longer clips, 4K output, and far more control over every shot.

AI is changing the skills kids will need in the future. Mark Rober’s CrunchLabs shows why curiosity, problem-solving and resilience matter.

Pollen Robotics’ Microduck is a tiny $399 biped designed to make physical AI training more accessible, letting developers experiment with new robotic behaviors on hardware that can survive the occasional tumble.

Claude now shares memory between chat and Cowork, letting users carry context across conversations while editing or deleting what it remembers.

A new Harvard-developed AI app helped visually impaired users complete an outdoor route faster and make less contact with obstacles in early testing.

An autonomous AI-guided drone has caused civilian deaths in Ukraine after selecting its own target, pushing the debate over AI weapons into far more serious territory.

Apps like Carrier Pidge and Roost are embracing slow messaging as users look for a break from constant notifications and digital overload.

Researchers found consumer spy-device detectors provided only a small improvement in finding hidden cameras and trackers, while confusing alerts could create dangerous false reassurance.

New zinc-iodine battery tech survives 60,000 cycles, promising safer, longer-lasting home solar energy storage.

Meta's smart glasses are running into another roadblock, and this time your next movie night could be affected.

A new Pew Research Center study finds signs of AI authorship across roughly one-third of web pages published since ChatGPT launched.

Meta has launched a Mac app for its AI assistant with screen sharing, dictation, and productivity tools for creators and businesses

MIT researchers found that AI-generated images often cannot be traced to individual training images as models scale to larger datasets

UC San Diego researchers have developed cheap passive reflective tiles that could improve mmWave 5G coverage without requiring powered equipment.

Teens are already using AI for homework and much more. OpenAI’s new ChatGPT experience wants to make sure they’re actually learning from it — with some serious guardrails attached.

A new FCC rule bans foreign-made robots from the U.S. unless they're mostly built domestically.

A blurry TikTok clip of a fish coming out of a borewell led scientists straight to a brand new species living underground beneath the India-Nepal border. Meet Gangaichthys indonepalicus.