Full-text search across 70835 articles. Combine with topic and date filters; results sorted by relevance.
A training-free, CPU-only lexical prompt-compression pipeline reduced prompt tokens by a mean of 40.3% (sigma = 9.2) while holding BERTScore-F1 at 0.876 against original-prompt output, according to an arXiv paper (2609.1…
Researchers Alpay Ozkan and colleagues introduced a pixel-wise planarity prediction framework for monocular plane segmentation from a single RGB image, published as arXiv:2609.13246v1. The method adds a dedicated planari…
Researchers introduced TestHallVQA, a multi-image visual question answering benchmark built from scientific exams that combines document-level scale with human-examination difficulty, according to the arXiv paper 2609.13…
Researchers introduced BVB, a benchmark that evaluates agentic video understanding by having multimodal agents programmatically reconstruct videos in Blender rather than answer questions. The benchmark targets agents tha…
Cognous has introduced the Open Control Stack, a four-layer guardrail framework (Declare, Control, Replay, Evidence) designed to constrain AI agents built on LangChain. The system uses an Agent Action Manifest to define …
Microsoft AI, under CEO Mustafa Suleyman, published a code of conduct for its MAI models that explicitly rejects any claim of consciousness or rights for the systems and makes safety a prerequisite for development rather…
GLM-5.3-Flash matched the cyber-exploitation performance of Claude Mythos Preview on ExploitBench at roughly 6% of the cost, according to a September 2026 blog post by James Mann. Running with a 1 billion token budget pe…
US President Donald Trump rejected calls for AI guardrails, saying the industry needs a "STRONG AND SMART PRESIDENT" and that his administration proactively regulates AI, while Vice President JD Vance called regulatory p…
AI coding tools Claude, Crush, Qoder, Factory, OpenHands, Codex, Cursor, Cline, Amp, Windsurf and OpenCode are converging on project-level configuration for safety boundaries, but the agnostic-ai project reports no share…
Reward AI released OM-1 (Omnibody Model 1), a general-purpose manipulation policy trained entirely on human demonstrations captured with a 7-DoF wearable glove, with no teleoperation or on-robot data. The policy runs on …
Apple's macOS Golden Gate release includes code integration that lets users replace Siri with Claude or ChatGPT as the default voice assistant backend, according to MacRumors. The change marks a major distribution breakt…
Slowave, a GitHub project, has added an adaptive memory layer that stores codebase decision context across AI agent sessions, addressing the loss of context when agents open a new session or switch tools. The project sta…
Apple released iOS 27 and iOS 26.7, which together fix more than 100 security vulnerabilities, with 75 of the fixes shared across both updates. The iOS 27 update addresses over 100 flaws and iOS 26.7 addresses over 80, i…
Microsoft CEO Satya Nadella said any pursuit of superintelligence must be grounded in the principle that AI not helping humanity and under human control is not worth pursuing, and that AI benefits must be diffused broadl…
An updated open source benchmark suite shows GraphQL is more token-efficient than REST and other JSON API technologies for connecting agents to deployed services, according to the benchmark's new numbers. The same update…
A developer's census of the official Model Context Protocol registry found that its 101,219 raw entries correspond to only 31,309 distinct servers, a 3.2x inflation caused by the registry retaining every historical versi…
GitHub rolled out three configurable tiers for Copilot auto model selection — efficiency, balance, and intelligence — letting users weigh cost, quality, and response time per prompt. The feature is currently rolling out …
Microsoft AI published a code of conduct for its MAI models that prioritizes human control over autonomy and performance, with AI chief Mustafa Suleyman stating, "If it isn't safe we shouldn't build it." The rulebook rej…
Former Federal Trade Commission Chair Lina Khan said on Sunday that law enforcers already hold authority to charge companies and their CEOs for releasing "dangerous, unvetted, or defective products," arguing there is "no…
Laurie Voss argues that as the cost of writing code collapses, the cost of reviewing, fixing and operating it is following, leaving product definition and usability as the whole job. Voss writes that this cost is per pie…