IDC research finds that networking infrastructure is a critical bottleneck for enterprises moving agentic AI from pilot to production, with security concerns (32.6%), automation challenges (26.8%), and staff limitations …
Only 5% of AI prototypes make it to production, with 95% failing due to enterprise infrastructure and risk constraints. Google engineers at YouTube solved this by building a prototyping stack that decouples experimentati…
Google Cloud announced a method for administrators to analyze and govern Gemini Enterprise app usage at scale by routing telemetry to BigQuery. The approach enables deep-dive forensics, including profiling adoption patte…
Barkup-bench's Study AD confirms that its core LLM agent editing stack, previously validated only on weaker models, transfers successfully to the production-tier claude-opus-4.8, passing all pre-registered gates with nea…
Apple sued OpenAI, alleging former employee Chang Liu stole confidential data and used it to benefit OpenAI. The lawsuit highlights Apple's failure to revoke access or retrieve devices from Liu, who accessed Apple's syst…
Noma Labs discovered a prompt injection vulnerability in GitHub's Agentic Workflows that allows an attacker to trick the AI agent into leaking private repository contents via a public issue. By embedding hidden instructi…
A developer built a population health risk stratification pipeline for Medicare Advantage plans that scores members on clinical and financial risk, then tiers them into categories such as rising-risk, high-risk, and cata…
Researchers at Anthropic and independent developer Elie Bak measured the "j-space" of open-source language models, finding that steering vectors in middle layers causally influence output up to 30 tokens ahead, form earl…
Meta faces a lawsuit from 26 employees alleging its internal AI system unfairly targeted workers with medical conditions and disabilities for layoffs. The suit seeks to temporarily block the layoffs while the employees p…
PrismML has compressed a 27-billion-parameter AI model to under 4 GB, enabling it to run on an iPhone while retaining 90 percent of its original performance. Apple is reportedly testing the compression technology, which …
Rockchip's RK3588 single-board computers, starting at $75, can run local large language models but are limited by memory bandwidth rather than the advertised 6-TOPS NPU. Decode speeds range from 41.6 tokens per second fo…
The White House launched the Gold Eagle initiative to preemptively identify and patch vulnerabilities before AI systems can exploit them, uniting federal agencies and industry partners. The initiative aims to set a prece…
OpenAI is developing a screenless home AI speaker with a built-in camera and sensors, powered by its GPT-Live LLM, to serve as a personalized home companion. The device, expected to cost $200-$300, faces competition from…
Spotify has expanded its AI voice interface feature to allow Premium subscribers to interact with the app via voice or text, aiming to boost user engagement and retention. The move is seen as a strategic experiment to en…
Amazon's Kiro CLI coding agent now integrates with Arize Skills to provide observability and evaluation for agent-generated code changes, enabling teams to trace behavior, run experiments, and validate improvements befor…
YC-backed startup Coasty launched an API that lets developers run autonomous computer-use agents on managed machines, compose them into workflows, and drop to lower-level prediction primitives when needed. The API suppor…
Digibee reports that while its team uses Claude Code for development, coding agents consistently fail at enterprise integration tasks due to three key limitations: they start from scratch each time instead of using pre-b…
Moki, a new AI tool, helps users quickly evaluate ideas by scoring them on six weighted factors, pulling live Google search demand, and identifying the cheapest experiment to test the biggest unknown. The tool aims to re…
Quickchat AI published a no-code guide for making AI agents' actions reliable, focusing on two settings—'Run only when' and 'Save to memory'—that gate API calls and preserve values across steps. The guide demonstrates th…
PrismML, a Caltech spinout, compressed Alibaba's 27-billion-parameter Qwen3.6-27B AI model to under 4 gigabytes using 1-bit and ternary quantization, enabling it to run natively on an iPhone 17 Pro at 11 tokens per secon…