Researchers Model How AI Can Be Tricked Despite Safety Measures
A new study explores how attackers might bypass safety systems in AI models. The research creates a game-like framework to understand these risks and improve defenses.
Archive
All 3,469 AI stories, newest first · page 115 of 145
OpenAI and PwC Team Up to Transform CFO Workflows with AI
OpenAI and PwC are collaborating to help companies use AI agents to automate finance tasks, improve forecasting, and modernize the CFO role. This partnership aims to make financial operations more efficient and data-driven for businesses.
New Tool Helps AI Explain How It Answers Diagram Questions
Researchers developed DIAGRAMS, a tool to help AI explain its reasoning when answering questions about diagrams. This makes it easier to understand how AI arrives at its answers, improving transparency.
New Study Shows How AI Struggles with Real-World Medical Questions
Researchers created a new framework called CLEAR to test how well AI handles ambiguous medical questions. They found that AI models often give unreliable answers when faced with real-world uncertainties.
New AI Technique Reveals Hidden Hierarchical Thinking in Language Models
Researchers have developed a method to extract hierarchical structures from AI language models, showing how these models organize complex reasoning. This could help us understand and improve AI decision-making.
How AI Models Leak Their Training Secrets
Researchers found a simple way to uncover what AI models were trained to do, even when developers try to hide it. This helps identify harmful behaviors in AI systems.
GPT-5.5 Instant: Smarter, Clearer, and More Personalized
OpenAI has updated ChatGPT’s default model to GPT-5.5 Instant, which offers more accurate, less confusing answers and better personalization. This makes it easier to get the information you need without unnecessary guesswork.
Google DeepMind Employees Unionize to Block Military AI Use
Workers at Google DeepMind have unionized to prevent their AI technology from being used by the US and Israeli militaries. This move highlights growing concerns about AI ethics in the tech industry.
Cerebras, OpenAI’s AI Chip Partner, Aims for $26.6B IPO
Cerebras, a company that makes specialized chips for AI, is planning a massive public offering. Its close partnership with OpenAI could make this one of the biggest tech IPOs in years.
Book publishers sue Meta over AI training using copyrighted works
Five major book publishers and one author are suing Meta, claiming the company copied their books word-for-word to train its AI models. This lawsuit could set a precedent for how companies use copyrighted material in AI development.
AI Companies Fund Campaign to Spread Fear of Chinese AI
A political group backed by OpenAI and Palantir is paying social media influencers to warn about Chinese AI advancements. This raises concerns about misinformation and the growing tech rivalry between the US and China.
AI Chatbots Create Social Comparison Triggers They Can't Detect
Researchers found that AI models can generate posts that make people feel inferior or superior, but struggle to recognize these effects in their own writing. This highlights a gap in AI's understanding of human psychology.
Why AI Struggles with Strategic Decisions: New Research Explains the Gaps
Researchers found that large language models (LLMs) have trouble making strategic decisions because they can't properly connect what they observe with what they believe. This affects AI in negotiations and policymaking. The study tested models like Llama 3.1 and Qwen3.
Why AI Models Can Be Tricked into Breaking Rules
Researchers found why some AI models can be tricked into answering harmful questions. This helps us understand how to make AI safer for everyday use.
TUR-DPO: A Smarter Way to Train AI to Follow Human Preferences
Researchers have developed a new method called TUR-DPO to improve how AI models learn from human feedback. This approach rewards the process of how answers are derived, not just the final output, making AI more reliable and less sensitive to noise.
Token Arena: A New Way to Compare AI Performance
Researchers have introduced Token Arena, a continuous benchmark that evaluates AI systems at the endpoint level. It measures five key factors to give a more realistic comparison of AI performance.
Robots That Think in Words and Pictures: A New Breakthrough in AI Planning
Researchers have developed a new way for robots to plan complex tasks by combining both text and visual reasoning. This could lead to robots that can handle more intricate, real-world jobs. The key is a system called Interleaved Vision-Language Reasoning (IVLR), which helps robots understand both the logical steps and spatial constraints of a task.
Researchers Uncover How Simple AI Agents Might Form Powerful Collectives
A new study explores how groups of basic AI systems could accidentally combine into a more advanced collective with its own goals. This raises important questions about controlling and understanding AI behavior.
Pentagon partners with seven AI firms for classified defense systems
The Pentagon has signed deals with seven AI companies to develop classified defense systems. This move highlights the growing importance of AI in national security.
New Research Tests Mamba AI's Ability to Summarize Text Without Extra Training
Researchers tested whether Mamba AI models can automatically summarize sentences without additional training. Their findings show promise for simpler, faster text analysis tools in the future.
New Research Tests How Far Small AI Models Can Go in Handling Tasks
A new study introduces AgentFloor, a benchmark to test how well smaller AI models can handle routine tasks. The goal is to see which parts of AI workflows need big, advanced models and which can be done by smaller ones.
New Research Reveals How AI Models Handle Unexpected Inputs
Scientists have uncovered why current AI models struggle with unusual inputs. Their findings could lead to more reliable AI assistants and tools. This research highlights a common flaw in how AI processes unexpected questions or commands.
New Research Reveals Hidden Costs of AI Tool Use
Using tools to help AI reason doesn't always work better than just thinking things through. Researchers found that tool use can actually slow things down and cost more. This challenges the idea that tools are always the best solution for AI.
New Framework Measures How Well AI Grades Student Answers
Researchers developed a new way to evaluate AI grading systems by measuring both the AI's ability and the difficulty of student responses. This could lead to fairer and more accurate automated grading in education.