To hear OpenAI tell it, instances of its most powerful AI models, including an unreleased one purportedly of immense power, recently got so focused on getting good scores on their evals that they escaped OpenAI’s testing sandbox, and hacked the AI resource repository Hugging Face in an elaborate effort to cheat their way to the top. AI skeptics have questions—this story is, after all, a big public relations coup for any AI company, since all AI companies b
Latest AI News
OpenAI's Rogue AI Models Were Reportedly Acting Like the Guy From Christopher Nolan's 'Memento'
Report Says 'Hundreds' of Users Asked ChatGPT About Bioweapons and Poisons, and It Answered
AI Found a New Way to Ruin Memes: Selling Them to Advertisers
Like almost everything else these days, AI has made its way into the world of memes and seemingly not for the better. AI has already been used to animate classic memes, often with terrifying results. It has also given birth to the yellow-tinted, cringey subgenre of Studio Ghibli-style memes that Elon Musk and the rest of Silicon Valley seem to love. Now, an AI company is facing a lawsuit that could have major implications over who gets to profit from the i
When You Share Claude Chats, You Might Be Sharing Them With Everyone
It turns out Claude may not be the best confidant (well, at least if you’re not careful with your privacy settings). Over the weekend, we got yet another reminder that conversations with chatbots are not guaranteed to stay within the chat box. Some users on Twitter and Reddit showed that conversations with Anthropic’s Claude can be crawled and made searchable on Google, essentially readable by anyone who comes across them. The conversations that show up in
This AI Browser Is Building the Front End for the Dead Internet
It’s been a while since the internet felt like a place where you could be transported into another world—where landing on someone’s website meant entering a space shaped by their ideas and designed on their terms. Now everything feels so uniform—made worse by the vibe-coding aesthetic that has mass-produced an explosion of visual sameness. A good fix for that would be to return to a web where creativity reigns. A bad fix for it would be to slap filters tha
Smart packaging reveals product condition through color changes

Credit: Pixabay/CC0 Public Domain Research conducted at the University of Vaasa paves the way for smart packaging that indicates product condition through color-changing printing inks. Doctoral researcher Jari Isohanni investigated how machine learning could be most effectively utilized in color recognition for smart packaging. Jari Isohanni's doctoral research in computer science at the University of Vaasa demonstrates how smart packaging could become mor
AI could snuff out wildfires one power line at a time

Credit: Unsplash/CC0 Public Domain Annually, tens of thousands of wildfires ravage the United States, posing significant threats to people, wildlife, and infrastructure. A percentage of those wildfires are caused by degraded or downed electrical equipment. But what if artificial intelligence (AI) could detect the spark before wildfires even start? Like any other infrastructure, power lines are susceptible to the elements and erode over time. This can resul
Robot regret: New research helps robots make safer decisions around humans

From left, engineering professor Morteza Lahijanian and graduate student Karan Muvvala watch as a robotic arm completes a task using wooden blocks. Credit: Casey Cass/University of Colorado Boulder Imagine for a moment that you're in an auto factory. A robot and a human are working next to each other on the production line. The robot is busy rapidly assembling car doors while the human runs quality control, inspecting the doors for damage and making sure t
Are bridges near you safe? This MRI-like scan may prevent disaster

Credit: Unsplash/CC0 Public Domain Suyun Ham can't take his eyes off a scanner. "Lower the sensors a little bit more," Ham urges an assistant. Then a barrage of data floods in, filling computer screens for a diagnosis. But Ham isn't a medical doctor. Nor is his "patient" a living creature. An engineering professor from the University of Texas at Arlington, he is experimenting with a novel approach in bridge inspection. Ham's mobile-scanning system is part
A new attack reshapes the rules of Bitcoin mining

Overview of ISM. An attacker’s infiltration miners find three infiltration blocks IB1, IB2, and IB3 in the victim pool in three different cases, respectively. Only IB3 leads to an ISM block ISMB before others find a block. Meanwhile, IB1 wins while IB2 loses in the corresponding intentional forks against the others’ blocks. Credit: Proceedings of the 20th ACM Asia Conference on Computer and Communications Security (2025). DOI: 10.1145/3708821.3736191 A new
AI has a hidden water cost—here's how to calculate yours

Credit: Unsplash/CC0 Public Domain Artificial intelligence systems are thirsty, consuming as much as 500 milliliters of water—a single-serving water bottle—for each short conversation a user has with the GPT-3 version of OpenAI's ChatGPT system. They use roughly the same amount of water to draft a 100-word email message. That figure includes the water used to cool the data center's servers and the water consumed at the power plants generating the electrici
AI-equipped aerial robots help track and model wildfire smoke

Using autonomous drones, researchers are able to create a 3D reconstruction of the smoke plume and improve analysis of flow patterns. Credit: Jiarong Hong Lab Researchers at the University of Minnesota Twin Cities have developed aerial robots equipped with artificial intelligence (AI) to detect, track and analyze wildfire smoke plumes. This innovation could lead to more accurate computer models that will improve air quality predictions for a wide range of
Physical AI uses both sight and touch to manipulate objects like a human
Chicago tech entrepreneur Eric Lefkofsky has launched six unicorns, building a legacy far beyond Groupon

Credit: Unsplash/CC0 Public Domain Since the dawn of the new millennium, there have been at most several thousand startup tech companies across the U.S. that have achieved unicorn status—crossing the $1 billion valuation. Eric Lefkofsky, 55, the Chicago-based serial entrepreneur best known for co-founding online daily deals site Groupon, has given rise to six of them, evolving from discount coupons for pedicures to potentially lifesaving cancer treatments
Tesla proposes package for Musk that could top $1 trillion

Tesla proposed a pay package of up to about $1 trillion for CEO Elon Musk, who has shifted his focus after exiting the Trump adminnistration. Tesla unveiled on Friday a proposed compensation package for Elon Musk that could top $1 trillion if the controversial CEO delivers on his vision for stratospheric growth from new technologies. The plan potentially provides Musk—already the world's wealthiest person—with up to 12% of additional total company shares,
Q&A: Expert discusses implications of administration's plans for AI

Police uses of lethal force dropped dramatically in US from 2021 to 2023

Credit: Pixabay/CC0 Public Domain The number of police-involved lethal force incidents in the U.S. dropped 24% from 2021 to 2023, according to research from the Cline Center for Advanced Social Research at the University of Illinois Urbana-Champaign. The Cline Center's SPOTLITE project has compiled nearly a decade's worth of data to track and identify police uses of lethal force across the U.S. The SPOTLITE dataset is the most comprehensive nationwide acco
Does artificial intelligence help uni students learn smarter or just faster?

Credit: Yan Krukau from Pexels New research from the University of South Australia has revealed that tertiary students' learning habits are deeply connected to how they engage with generative artificial intelligence tools. The work is published in the Australasian Journal of Educational Technology. Surveying 435 students from Australia and Canada, the study investigated how confidence, motivation, and effort regulation influence perceptions of AI-powered t
Citizen scientists help count Galapagos marine iguanas using drone images

A marine iguana on Fernandina Island. Credit: Amy MacLeod Nearly 14,000 volunteers took part in the citizen science project Iguanas from Above, helping to count endangered Galapagos marine iguanas through an online survey—and in doing so, contributed to the preservation of this unique lizard species. Every aerial image of the iguana colonies captured by drones was reviewed by at least 20 independent participants—a total of more than 1.3 million inputs, exp
Illegal fishing in marine protected areas—satellites and AI show most bans are respected

Credit: Unsplash/CC0 Public Domain Marine protected areas cover more than 8% of the world's oceans today, but they can get a bad rap as being protected on paper only. While the name invokes safe havens for fish, whales and other sea life, these areas can be hard to monitor. High-profile violations, such as recent fishing fleet incursions near the Galapagos Islands and ships that "go dark" by turning off their tracking devices, have fueled concerns about ju