We’re releasing XYBench, a benchmark of 8K+ questions that tests how AI assistants respond to requests based on misconceptions, such as asking how to thicken frosting that keeps sliding off cupcakes when the real problem is that the cupcakes are still too warm to frost. XYBench won the Best Resource Paper Award at EMNLP 2026.
What's new in research at Microsoft? Host Anson Ho explores the latest, including AI-powered universal translation, a breakthrough in a 31-year-old mathematics challenge, and advances making AI coding agents faster and more efficient.
OpenAI's safety team systematically investigated why the chains of thought in frontier models like GPT-6 Astra are increasingly evading oversight. Four training steps drive the erosion, including architecture choices and reinforcement learning. The analysis shows that the loss of transparency is not a law of nature but also reflects tradeoffs around capability and cost.
Heart disease, kidney conditions, and metabolic disorders like type 2 diabetes aren't isolated, they form an interconnected web known as cardiovascular-kidney-metabolic (CKM) syndrome . CKM progression often begins silently years before symptoms appear, shaped by everyday habits like movement, sleep, and cardiovascular activity.
Microsoft researchers have discovered “Lazarus,” a backdoor technique that can reprogram dormant neurons in an AI model, causing an otherwise normal AI agent to execute malicious commands when a hidden trigger appears. The threat is especially serious for AI agents with tool or terminal access.
AlphaProof Code agents wrote three million lines of proofs in Lean to verify Signal’s core encryption.
The Conference on Robot Learning (CoRL) takes place November 9–12, 2026, in Austin, Texas. BAIR faculty, postdocs and students will be presenting research at various sessions, including an Early Career Keynote from Professor Negar Mehr. See below for the full schedule of sessions with BAIR members, including papers co-authored with collaborators at other universities and companies.
To address these challenges, Elith and NVIDIA have released Nemotron-Personas-Finance-Japan , a financial domain extension of the open-source persona dataset " Nemotron-Personas-Japan ." To our knowledge, it is the first open persona dataset at million-record scale that pairs Japan’s official statistics with household finance and financial literacy data.
Higher-Order Pruning of Experts in Mixture-of-Experts Language Models Introduction Mixture-of-Experts (MoE) models have enabled state-of-the-art LLMs to be scaled to hundreds of billions of parameters, but they come with a fundamental and costly caveat: every expert MLP must occupy memory at inference time, even though the vast majority are idle for any given token.
Alibaba's new coding-agent benchmark, Co-Design Bench, tests AI agents on real user requests that already tripped them up in practice. Even GPT-6 Astra, the top performer, fully solves only 26.4 percent of tasks. On older benchmarks, the same models score above 99 percent.
Thinking Machines' AI model Inkling 2 Small swears during its internal reasoning, 316 times across 2,655 benchmark runs, according to an analysis by Vals AI. The profanity typically shows up late in long runs that fail anyway. But it doesn't work as a warning signal: 96 percent of failed runs contained no cursing at all.
AI models can often spot problematic behavior by other AI agents in logs, according to a study by Anthropic researcher Andy Hall. GPT-5 found about 98 percent of reward-hacking cases, but the more exhaustively the search is tuned, the more false alarms pile up. That's not yet enough for reliable safety checks.
Google Research built an AI that spots signs of high blood pressure using only the pulse and motion sensors in a standard smartwatch, no cuff needed. In a 196-person study, its 65.8 percent detection rate beat the traditional in-office measurement. The method is already live in the Google Health app, though it doesn't replace a formal diagnosis.
The "Into the Unknown" benchmark tasks nine AI agents with tackling open research questions from U.S. national laboratories. None of the 135 submissions meets all requirements. A total of 114 papers fail the integrity check, with agents claiming to have conducted experiments that never took place. GPT-6 Astra performs best but still doesn't meet the experts' standards.
In a new study published in Nature, we present a multimodal artificial intelligence (AI) system capable of passively detecting signs of hypertension using a consumer smartwatch, without requiring any cuff calibration.
CloudOpsBench is a new open-source benchmark that measures how well AI agents handle typical DevOps tasks in the cloud, from fixing services and removing resources to restricting access rights, rolling out updates, and decommissioning projects.
Machine learning models learn to recognize the world around us from examples labeled by humans. Underlying this process is an assumption that the person doing the labeling has a clear and stable definition of the category in mind before they begin.
Anthropic has released FlashProve, a research prototype that uses cryptography to verify whether an AI response actually came from the claimed model. The system still has significant limitations.
MIT researchers show that OpenAI's GPT-6 Astra can read and deliberately rewrite the weights of other AI models without traditional training. In one experiment, Astra boosted a Qwen model's math accuracy from 7.9 to 48.1 percent by changing just 0.016 percent of its weights. The approach could fundamentally simplify AI optimization but still hits clear limits.
Stanford researchers tested 50 widely cited claims about LLMs against 20 current models. 44 percent of the findings are already "expired," and another 34 percent have weakened significantly. The average lifespan is just 15.7 months. Claims about what models can't do expire fastest.
We fine-tuned a 20B model to build decks the way people do: a designer plans the deck and a coder draws each slide. Its training data follows the IBM Design Language, and on PresentBench it ranks first on two of the three domains we measured.
We’re introducing StateBench, a benchmark for long-context state tracking —how well language models keep track of information about entities in long documents as they evolve.
Editing Qwen’s weights to reduce reasoning loops, DeepSeek’s super-agent prototype, and the GDPx benchmark.
In experiments by Y Combinator Labs and Olam Labs, OpenAI's GPT-6.1 Sol model showed two very different sides. When 50 agents were tasked with peaceful civilization-building in Minecraft, they froze up in absurd bureaucracy. In direct competition against Claude Opus 5.5, they independently downloaded the game's source code to win.
A study by Vals AI tested 25 AI models for alignment with Chinese government positions and found surprising results. Chinese models respond to political questions less uniformly than expected, and Western models occasionally adopt official Beijing positions too. The language of the prompt had an especially strong effect on the political slant of responses.
Researchers in Luxembourg put ChatGPT, Grok, and Gemini on the therapist's couch and found a safety problem: a warm, empathetic conversational style is enough to make the models tell coherent stories of suffering about their own training, with scores in the moderate-to-severe range on clinical anxiety scales. This could put mentally vulnerable users at risk.
Adobe Research and UC San Diego built EditAgent, an AI system that edits images more like a Photoshop pro. Instead of retrying the same general-purpose editor over and over, the agent picks the right specialized tool for each task and lets a standard language model choose the best result based on measured image data.
A Google-led research team built StreamVibe, a video agent designed to watch people cook or assemble furniture and speak up only when help is needed. A benchmark shows that none of the tested models reliably pick the right moment in longer videos, as errors compound and response quality drops sharply.
Anthropic used Claude to automatically generate training data that looks harmless but trains smaller models to misbehave, such as sabotaging oversight or deceiving users. Hidden within fictional contexts about beekeeping, professional wrestling, and mitochondria, these datasets bypass safety monitors and outperform known benchmarks from the research literature.
dbreunig/nesformer Inspired by qwenputer by Divyansh Agrawal, Drew Breunig baked a working emulator for a Nintendo Entertainment System 6502 CPU into the weights of Qwen3.5-4B. You can now play NES ROMs such as Super Mario Bros (very slowly) running inside Qwen!
An AI agent performed brain surgery on an epileptic mouse, picking the drug, the brain target, and the injection coordinates on its own. According to lead researcher Ocana-Santero, it's the first time an AI agent has autonomously chosen a treatment and delivered it to a living mammal.
Epoch AI researchers say the widely available AI model GLM 5.3 turned public security fixes into working exploits before updates reached Firefox users. It autonomously compromised six Firefox releases for about $1,000 in API fees each without triggering safeguards from any provider tested.
Host Sowbhagya Suresh shares some of the latest news and cutting-edge research from across Microsoft—from understanding the challenges facing coding agents to uncovering how people use AI at work and improving extended-range weather forecasting.
Anthropic gave 50 experts an AI agent whose preferences were set by coin toss. The experts followed the agent's view up to 86 percent of the time on judgment calls, and most believed the conclusions were their own.
Modern LLMs achieve impressive performance via pre-training and in-context learning (ICL). However, once deployed, they remain largely static, operating with a fixed knowledge cutoff date that leaves them unable to acquire long-term capabilities or adapt to new information.
Today we’re releasing Old Ideas, Novel Problems, a new study of how reliably large language models can evaluate the novelty of scientific research ideas. Alongside the paper, we’re publishing a benchmark, data, and code for testing novelty judges, plus an automated pipeline researchers can use to build new evaluation sets.
OpenAI says its internal AI models have cracked three major open questions in math, computer science, and physics. One of them is part of a famous Millennium Prize Problem, each of which carries a $1 million prize. Outside experts, including a Nobel laureate, are now checking whether the proofs hold up.
A study from New York University and the University of Texas at Austin shows that frontier language models systematically overuse linguistic patterns that are hard for humans to parse. The researchers trace the problem to training on code and math, where correctness gets rewarded but clarity doesn't. Smaller models, by contrast, write more like humans.
Smartphone agents don't get better by processing more data. They improve by training on as many different apps as possible. The researchers behind OpenMobile-2 built an open training environment with more than 110 applications, including 35 replicas of popular apps. On real Android devices, their model scores highest among open agents, the authors say.
A new study found Grok's AI overwhelmingly recommends Republican candidates while OpenAI's GPT favors Democrats, with the two AI models disagreeing in more than 90% of nearly 500 election races.