GitHub's new ModernBERT-based classifier evaluates candidate secrets in under 2ms, potentially doubling the number of leaked credentials blocked before entering repository history.
Epoch AI tested six frontier models on real tasks from its own research team, finding they handle well-defined work but fail at judgment-heavy, open-ended tasks.
Goodfire's probe-plus-judge cascade watches Kimi K3 and GLM 5.3 activations in real time, matching LLM judges at 50x lower cost while blocking universal jailbreaks.
Anthropic pledges $150M over three years to put Claude, Claude Code, and API credits in the hands of scientists at NASA, NIH, NSF and other federal labs.
Canada's lone frontier model lab plants a flag in the capital, bringing engineers and solutions architects within walking distance of its biggest customer: the federal government.
Intelligent Internet released Evoke, a Postgres extension that folds sparse semantic retrieval into the BM25 inverted index, no vector store or GPU required.
IBM's 1M-parameter TSPulse foundation model handles classification, anomaly detection, imputation, and search with GPU-free inference and open weights.