All news

Former GitHub CEO Thomas Dohmke launched Entire, a code platform built around review for AI coding agents, arguing review is now the main bottleneck.

Multi-token prediction and alternative tokenization help LMs learn latent graph shortest-path search.

http2shell is a new RCE affecting Node, Python and other stacks.

Mix-attention fuses multimodal LLM features into sequential recommendation.

HPE patches 72 ArubaOS flaws, eight max severity RCE or auth bypass.

Open LLMs on the Spanish MIR medical exam with uncertainty and image use.

Korean police link a gang that breached 90+ organizations and stole 8.2M records to the ARTEX AI autonomous pentesting framework.

Cloudflare launched a closed beta key manager that keeps decryption keys in chosen regions and protects them with post-quantum cryptography.

LLM-aided RTL timing optimization via logic metamorphosis.

Microsoft patches CVE-2026-111497, an out-of-bounds write in the Brotli decoder, across supported .NET versions.

Audits the hand-written deny sets and registries behind learned access control.

Action learning for controlling scientific simulations.

Neural transistor models speed up SPICE and cell characterization.

Gram 3.4.0 updates rustls-webpki and webpki-roots after the eMudhra CA incident.

Prompting strategy effects on LLM judges vary by judge model and language.

Apple blocks the compromised eMudhra intermediate CAs since October 19 and explains what Terminal users must update.

Particle physics foundation models for track finding.

Evaluates the non-generative Jev model as a relevance judge and reranker for Portuguese IR.

Arista patches an exploited zero-day in VeloCloud Orchestrator.

Microsoft says China-aligned attackers used AI agents and stolen ConnectWise credentials to hit 2,250 cloud tenants.

EcoxPU orchestrates disaggregated LLM inference across heterogeneous accelerators with agents to cut energy use.

DWI speeds up on-device LLM decoding by picking the width of a nested model per step.

Shows reward model scores are not comparable across prompts.

Neural approximation for reachability of vision-controlled systems.

Dream-1 robot foundation model works out of the box on a bimanual robot via API.

Survey and audit of recommender research where LLM agents edit recommender configs or code and retrain them.

Text-query-guided video summarization with a self-attentive conv network.

Multimodal semantic fusion for video recommendation.

Trains small LMs to select tools across corpora of over ten thousand tools.

MATCHA is a Chinese social media dataset covering several mental health disorders, with cross-lingual analysis.