The edition highlights a proof of the KLS conjecture and a new method, APEC, that reduces coding agent processing costs by up to 25.4%.
The attack requires no retraining and passes standard safety evaluations, with triggers as simple as a date change.
A research prototype from Microsoft Research and GitHub Next that shifts review from line-by-line code to plain-language intent, using formal proofs to ensure agent-written Rust code matches specifications.
Walker argues that the 2013-2016 Cyber Grand Challenge relied on expert systems rather than machine learning, and that current AI capabilities have disrupted the traditional balance of offensive scarcity.
The platform uses publicly available data sources like O*NET and SEC filings to generate realistic test scenarios without using customer data.
BioEmu2, a foundation model for biomolecular structure, is now available in public preview, while new frameworks improve coding agents and weather forecasts.
The system uses a high-level language and compiler to reduce the cost and complexity of creating interactive visualizations with AI agents.
Researchers proved the 1995 Kannan-Lovász-Simonovits conjecture, establishing a dimension-free geometric principle for log-concave distributions.
Magnetar proves RISC-V execution at 7.2 to 8.7 million instructions per second on a laptop using only standard hash functions.
StudentSim outperforms baselines like GPT-5.4 and Maia2 in chess, writing, and math simulations, achieving 90.5% tutor accuracy in reinforcement learning tests.
Microsoft Research Asia released Agent Lightning v1.0, a 3,500-line framework that raised Qwen3.5-9B's Pass@1 on SWE-bench Verified by 14.6 percentage points using only 6,000 training samples.
Microsoft Research scientists Jennifer Neville and Chad Atalla discuss how AI models degrade in multi-turn interactions despite high single-turn benchmark scores, highlighting the need for new evaluation methods and user strategies.
The system provides 30 to 60 minutes of advance warning and detected nearly 80% of major space-weather events during evaluation.
Quine combines a multimodal world model with an interactive harness to prioritize therapeutic compounds, validated in collaboration with the Broad Institute.
The lab has launched nine new projects with NUS and NTU and engaged over 300 students through its summer school since opening in July 2025.
MS unveiled Physical AI Toolchain features that offload robotics inference to the edge or cloud, extending battery life by up to 160% and improving performance.
Microsoft has open-sourced the code and weights of RetroChimera, a retrosynthesis AI published in Nature, under the MIT license.
Chris White leads AI security and infrastructure innovation at Microsoft Catalyst Lab after developing dark web search tools at DARPA.
Flash models maximizing the efficiency of GigaPath and GigaTIME have been released, enhancing accessibility for large-scale pathology research.
Microsoft Research released Skala 1.1 with improved accuracy and expanded integration with major simulation codes.
MindTopo measured the gap between VLMs' static topological recognition and interaction planning capabilities.
CARE-X is presented as a chest X-ray VLM that combines generation and structured prediction.
Microsoft has released Orchard, an open-source framework for scaling agent training and evaluation.
Microsoft has developed Echoverse, a highly precise training environment designed to maximize the performance of computer-use agents.
AI Briefing