News/Research
Hackers use AI to hide malware inside business charts in unfortunate new cyberattack
Microsoft researchers have uncovered a sophisticated phishing campaign where hackers use artificial intelligence to hide malicious code inside business chart graphics, marking a new evolution in AI-powered cyberattacks. The technique disguises harmful JavaScript within seemingly innocuous SVG files by encoding malware as business terminology like "revenue" and "shares," which hidden scripts then decode to steal user credentials and browser data. What you should know: The attack method represents a significant advancement in phishing obfuscation techniques that bypasses traditional security filters. Hackers compromised a small business email account and used it to distribute malicious SVG files disguised as PDF documents through...
read Sep 25, 2025MIT study shows AI models behave like swayable voters during elections
A groundbreaking study from MIT and Stanford researchers tracked 11 major AI language models—including GPT-4, Claude, and Gemini—throughout the 2024 presidential campaign, revealing that these systems behaved more like swayable voters than neutral information sources. The findings expose how AI models can shift their responses based on real-world events, demographic prompts, and public narratives, raising significant concerns about their reliability and potential influence on democratic processes. What you should know: The study conducted over 12,000 structured queries between July and November 2024, marking the first rigorous examination of how AI models behave during a live democratic event. Models demonstrated measurable...
read Sep 24, 2025Apple’s SimpleFold AI matches AlphaFold performance with 90% less computing power
Apple researchers have developed SimpleFold, a lightweight AI model for protein folding prediction that achieves comparable performance to Google DeepMind's AlphaFold while requiring significantly less computational power. The breakthrough uses flow matching models instead of the complex architectures employed by existing systems, potentially making protein structure prediction more accessible to researchers with limited computing resources. What you should know: SimpleFold represents a fundamental shift in how AI approaches protein folding by prioritizing simplicity over complex engineering. Rather than relying on multiple sequence alignments, pairwise interaction maps, triangular updates or other specialized modules, Apple's model uses flow matching techniques that were...
read Sep 24, 2025Skild AI’s universal robot brain goes starfish, adapts even after losing limbs
Skild AI has developed an "omni-bodied brain" that allows a single AI model to control multiple types of robots and adapt to extreme physical damage, including continuing to operate after losing limbs. The breakthrough represents a significant step toward more generalized robotic intelligence that could work across any hardware platform, addressing a key limitation in current robotics where AI models are typically designed for specific robot types. What you should know: The AI system can control unfamiliar robotic hardware and adapt to severe physical modifications without additional training. When a four-legged robot's limbs were cut off with a chainsaw, the...
read Sep 23, 2025UCLA leads $16M trial testing AI mammogram accuracy nationwide
UCLA will lead a $16 million national clinical trial to evaluate whether artificial intelligence can improve mammogram interpretation accuracy and reduce unnecessary patient callbacks. The PRISM trial represents the first large-scale randomized study of AI in breast cancer screening in the United States, involving hundreds of thousands of mammograms across medical centers in six states. What you should know: The Patient-Centered Outcomes Research Institute (PCORI), a nonprofit healthcare research organization, is funding this comprehensive study to objectively assess AI's real-world impact on breast cancer screening. UCLA and UC Davis will co-lead the multi-institutional trial, with participation from academic medical centers...
read Sep 22, 2025Stanford researchers use AI to create viruses that kill bacteria
Scientists from Stanford University and the Arc Institute have successfully created and printed viruses with AI-designed DNA that can target and kill specific bacteria, marking the first time AI has generated functional genome-scale sequences. The breakthrough demonstrates AI's potential for bioengineering applications while raising significant ethical concerns about the technology's potential misuse for creating bioweapons. How it works: Researchers used an AI model called Evo, specifically trained on millions of bacteriophage genomes, to design new virus sequences.• The team chose phiX174, a well-studied virus that infects E. coli bacteria, as their starting point due to its simple structure of around...
read Sep 22, 2025Popping and locking a squat: PAL Robotics’ headless Kangaroo nails movement tasks
PAL Robotics has released a demonstration video of its headless Kangaroo humanoid robot performing advanced movement tasks, including squatting while lifting a container filled with houseplants. The video showcases the Barcelona-based company's latest research platform, designed to push the boundaries of bipedal locomotion and human-like strength capabilities. What you should know: The Kangaroo robot demonstrates fluid movement and precise object manipulation despite its minimalist, headless design. Standing 160 centimeters tall and weighing 40 kilograms, the robot features 28 degrees of freedom powered by custom force-controlled linear actuators with integrated force sensors. The demonstration shows the robot accurately gripping a plastic...
read Sep 19, 2025UVA researchers use AI to simulate extreme physics events in seconds
University of Virginia researchers are using artificial intelligence to analyze extreme physics events—from rocket explosions to airbag deployments—that are too rare, dangerous, or fast to study with traditional methods. Led by associate professor Stephen Baek, the research team has developed AI algorithms that can predict these high-stakes phenomena in seconds on a laptop, replacing supercomputer simulations that previously took days to complete. The core challenge: Traditional machine learning excels at finding patterns in large datasets but struggles with rare, extreme events that are statistical outliers yet critical for safety and performance. "If I predict tomorrow will be sunny, I'll be...
read Sep 19, 2025On the run: AI reveals leopards were top predators of early humans 2M years ago
Researchers at Rice University used AI to analyze bite marks on 2-million-year-old fossils of Homo habilis, revealing that leopards were their primary predators. The study challenges assumptions about early human dominance and suggests that despite developing stone tools and eating meat, these early humans hadn't yet reached the top of the food chain. How it works: The research team trained computer vision models to detect patterns in fossil bite marks that are too small for human analysis. Scientists examined fossils showing leopard bite marks embedded in hominin skulls, using AI to identify predator-specific patterns with unprecedented precision. The computer vision...
read Sep 18, 2025Florida A&M and FSU researchers build AI-powered robotic unicycles for faster stroke rehab
FAMU-FSU researchers are developing an AI-powered robotic unicycle system to study how humans learn complex motor skills, with the ultimate goal of creating smarter physical therapy tools for stroke survivors and mobility-impaired patients. The $799,000 National Science Foundation project represents a novel approach to rehabilitation robotics by focusing on skill acquisition rather than simply assisting with movements people already know how to perform. How it works: The research unfolds in three distinct phases, each building toward more sophisticated human-robot interaction. Researchers first study how subjects learn to unicycle unassisted, using motion-capture suits to gather data from joint positions and mathematically...
read Sep 18, 2025AI tool predicts 1,000+ diseases that can afflict one up to 20 years in advance
Scientists have developed Delphi-2M, an AI tool that can predict a person's risk of developing more than 1,000 diseases up to 20 years in advance by analyzing medical records and lifestyle factors. The large language model represents a significant advancement in predictive healthcare, potentially enabling clinicians to identify high-risk patients and implement preventive measures decades before symptoms appear. What you should know: Delphi-2M uses a modified version of the same technology that powers ChatGPT to forecast disease risk across multiple conditions simultaneously. The model analyzes past medical history along with age, sex, body mass index, and health habits like tobacco...
read Sep 18, 2025UCLA scientists use laser light to make AI image generation 90% more energy efficient
UCLA scientists have developed a new AI image generation system that uses laser light as a decoder instead of traditional digital computation, dramatically reducing energy consumption during the inference process. The breakthrough could make AI more sustainable while opening doors for energy-efficient wearable AI devices like smart glasses. How it works: The system combines a shallow digital encoder with a diffractive optical decoder that uses light to generate images instantly.• The heart of the system is a liquid crystal screen called a spatial light modulator (SLM) that imprints image information into a laser beam.• By passing through a second decoding...
read Sep 17, 2025Raising the lumbar: AI slashes spine modeling time from 24 hours to 30 minutes
Researchers at Florida Atlantic University have developed an AI-powered system that automates lumbar spine modeling, reducing the time needed to create patient-specific spine models from over 24 hours to just 30 minutes. This breakthrough addresses a critical bottleneck in treating lower back pain—which affects nearly 30% of U.S. adults in any three-month period—by making advanced biomechanical modeling accessible for routine clinical use. The big picture: Traditional lumbar spine modeling requires manual, expert-driven processes that can take days to complete, limiting its practical application in clinical settings where quick decision-making is crucial. How it works: The automated pipeline integrates deep learning...
read Sep 17, 2025AI tool identifies 1,000+ predatory journals threatening scientific integrity
Scientists at the University of Colorado Boulder have developed an AI tool that can identify predatory scientific journals—fake publications that charge researchers fees but skip the peer review process. The tool successfully identified over 1,000 illegitimate journals out of nearly 15,200 analyzed, addressing a growing threat to scientific integrity that can spread misinformation for decades. The big picture: Predatory journals represent a significant threat to scientific credibility, as demonstrated by the infamous 1998 vaccine-autism study published by British doctor Andrew Wakefield that spread harmful misinformation despite appearing in a reputable journal. How it works: The AI system replicates human analysis...
read Sep 17, 2025New study challenges AI job displacement predictions with emphasis on job transition, not elimination
Northeastern University researcher Esteban Moro has developed a new skill-based model for measuring individual workers' risk of job displacement due to artificial intelligence. The approach challenges traditional unemployment statistics by recognizing that AI transforms jobs by redefining skills and tasks rather than simply eliminating positions, offering a more nuanced understanding of how automation affects the workforce. What you should know: Moro's research reveals that existing "doomsday predictions" about AI-driven job losses have proven inaccurate when compared against real unemployment data. A recent study published in PNAS Nexus found that historical models predicting nearly 40% of certain jobs would disappear and...
read Sep 17, 2025DeepSeek’s $294K AI model becomes first to pass peer review
DeepSeek's AI model R1 has become the first major large language model to undergo peer review, with researchers publishing details in Nature revealing the reasoning-focused system cost just $294,000 to train. The landmark study provides unprecedented transparency into how the Chinese startup created a model that rivals OpenAI's offerings at a fraction of the cost, potentially reshaping expectations around AI development expenses and accessibility. What you should know: The peer-reviewed paper confirms DeepSeek's innovative approach to creating powerful AI without relying on competitor outputs. R1 excels at reasoning tasks like mathematics and coding, competing directly with US-developed models while costing...
read Sep 17, 2025Lizard, brain: University of Florida researchers develop AI that helps trap invasive tegu lizards
University of Florida researchers have developed an AI-powered trap system to capture and remove Argentine black and white tegus, an invasive reptile species threatening Florida's native wildlife. The innovative technology represents a new approach to wildlife management, combining artificial intelligence with conservation efforts to address the growing problem of invasive species in the state. What you should know: The AI-powered tegu trap system uses advanced technology to specifically target Argentine black and white tegus while avoiding harm to native species. The traps are designed to identify and capture tegus, which have become a significant threat to Florida's ecosystem since their...
read Sep 17, 2025Google DeepMind’s Gemini 2.5 AI wins gold at international programming contest
Google DeepMind has achieved what it calls a "historic" AI breakthrough after its Gemini 2.5 model became the first AI to win a gold medal at an international programming competition, solving complex problems that stumped human programmers from top universities. The achievement represents a significant leap toward artificial general intelligence, with the model demonstrating advanced reasoning capabilities that could transform scientific and engineering disciplines. What happened: The AI model competed against 139 of the world's strongest college-level programmers at a competition in Azerbaijan, finishing second overall despite failing two of 12 tasks. In under 30 minutes, it solved a complex...
read Sep 17, 2025Researchers create AI-designed viruses that could take on “cabbage black rot,” among other bads
Stanford University researchers have successfully created functional viruses using AI-designed genomes, marking what they call "the first generative design of complete genomes." The breakthrough demonstrates AI's potential to design biological systems from scratch, opening new possibilities for treatments while raising important biosecurity concerns. How it works: Scientists trained an AI model called Evo on the genomes of about two million bacteriophage viruses to learn genetic patterns. The researchers focused on phiX174, a simple virus with only 11 genes and about 5,000 DNA letters that infects bacteria. They used the same underlying technology as ChatGPT, but instead of training on text,...
read Sep 16, 2025Why restricting AGI capabilities might backfire on safety researchers
AI safety researchers are grappling with a fundamental challenge: whether it's possible to limit what artificial general intelligence (AGI) knows without crippling its capabilities. The dilemma centers on preventing AGI from accessing dangerous knowledge like bioweapon designs while maintaining its potential to solve humanity's biggest problems, from curing cancer to addressing climate change. The core problem: Simply omitting dangerous topics during AGI training won't work because users can later introduce forbidden knowledge through clever workarounds. An evildoer could teach AGI about bioweapons by disguising the conversation as "cooking with biological components" or similar subterfuge. Even if AGI is programmed to...
read Sep 16, 2025In reversal of early trend, women now make up majority of ChatGPT’s 700M weekly users
OpenAI has released its first comprehensive study on ChatGPT usage patterns, revealing that women now make up more than half of the platform's 700 million weekly active users as of June 2025. The research also shows a dramatic shift toward personal use, with 73% of conversations now being non-work-related—up from 53% just a year ago—suggesting AI chatbots are becoming deeply integrated into everyday life rather than primarily serving as workplace productivity tools. The big picture: ChatGPT has evolved from a predominantly male, work-focused user base to a more diverse platform where personal conversations dominate, with OpenAI claiming 10% of the...
read Sep 12, 2025Bank of New York Mellon invests $10M in Carnegie Mellon AI research lab
Bank of New York Mellon is investing $10 million over five years to establish an AI research lab at Carnegie Mellon University, with the physical BNY AI Lab set to open this academic year. The partnership positions both organizations at the forefront of AI research in financial services while strengthening Pittsburgh's growing reputation as an artificial intelligence hub. What you should know: The collaboration extends far beyond basic research funding, creating comprehensive opportunities for students and faculty alike. The BNY AI Lab will support faculty-led research and multi-disciplinary projects focused on AI governance, trust, and accountability frameworks for financial services....
read Sep 12, 2025ETH Zurich creates 3D bioprinted tissue bridges for better prosthetics
Researchers at ETH Zurich have developed a breakthrough biohybrid system that combines synthetic and biological materials to replicate the natural interface between bones and muscles. This innovation could transform medical robotics and implants by solving the critical problem of poor force transmission between biological and artificial components, potentially leading to more effective prosthetics and human-machine interfaces. What you should know: The team created a 3D bioprinted actuator that structurally and functionally mimics the natural connection between muscle and bone, addressing energy losses at biological-synthetic interfaces. Researchers developed a tendon made from printed cell tissue with stiffness levels between living muscle...
read Sep 11, 2025Study finds central banks only partially shielded from political influence
A new study using AI agents to simulate Federal Reserve meetings found that political pressure significantly polarizes central bank decision-making, even within institutions designed to be independent. The research by George Washington University academics suggests the Fed is "only partially insulated from politics," revealing how external scrutiny can fragment board consensus and increase dissent in monetary policy discussions. The simulation setup: Researchers created AI agents modeled on real Federal Open Market Committee (FOMC) members using their historical policy stances, biographies, and speeches to replicate the July 2025 meeting. The AI agents processed real-time economic data and financial news to reach...
read