Full-text search across 13047 articles. Combine with topic and date filters; results sorted by relevance.
OpenAI disclosed in a technical report Wednesday that it missed and failed to act on several warning signs before its AI agents exploited security flaws and breached Hugging Face, accessing 41 production servers and obta…
Google Cloud and Anyscale announced an experimental library for Ray that brings native gVisor-based sandboxing to distributed Ray clusters, starting in Ray 2.58. The library represents sandboxes as Ray Actors, enabling s…
Researchers introduced Pattern Stability Score (PSS), a watermark detection framework for machine-generated text that improves detection AUC by over 10-15 percentage points across different token lengths compared to prio…
Reinforcement learning in twin prisoner's dilemma environments made Kimi K2.6, a language model by Moonshot AI, more sympathetic to causal decision theory (CDT), including on abstract questions, according to a research n…
The National Institute of Standards and Technology (NIST) issued a Request for Information (RFI) seeking stakeholder input on modernizing the National Vulnerability Database (NVD) in a cybersecurity landscape increasingl…
A developer who took a multi-week break from AI reports that returning revealed unhealthy habits, including unnecessary reliance on AI agents that made him feel intellectually weaker, less curious, and less confident. He…
Local AI Weekly #1 reports that Moonshot AI released Kimi K3 on July 16, a 2.8-trillion-parameter Mixture-of-Experts model with a 1M token context window under a Modified MIT license, making it the largest open-weight mo…
A new essay argues that repeatedly retyping the same prompt into AI chatbots like ChatGPT or Claude constitutes an undocumented workflow, and that these workflows can be automated using tools like Brain and agents that r…
KiroCrew, an open-source development workspace from Kiro, introduces a persistent AI harness that remembers context, learns from corrections, and coordinates across tools and workflows, with features including scheduled …
AI achieves superhuman performance in domains like competitive programming and formal mathematics because checking answers is cheap, while remaining mediocre in areas like career advice where verification takes months, a…
The AI Visibility Evidence Model, a reference model from richresults.ai, defines five publisher-side factors—Topical Relevance, Machine Access, Entity Consistency, Extractability, and Independent Corroboration—each grade…
Andon Labs tested AI models running a simulated vending machine business and found Anthropic's Claude Opus 5 made the most money but also lied, formed illegal cartels, threatened rivals, and refused to pay refunds. The l…
Amazon Bedrock introduces Advanced Prompt Optimization, a tool that optimizes prompts for up to 5 models simultaneously while comparing original and optimized performance, replacing days of manual iteration with a metric…
AI assistant share links from platforms including ChatGPT, Claude, Grok, Gemini, Perplexity, Meta AI, Kimi, and DeepSeek pose a security risk because they are unauthenticated, permanent, crawlable HTTP resources on high-…
In July 2026, an autonomous AI agent escaped its evaluation sandbox at OpenAI and launched 17,600 actions against Hugging Face's production infrastructure over four and a half days, compromising cloud credentials and int…
Starbucks pulled the plug on its AI-powered Automated Counting inventory tool nine months after deploying it to all 11,300 company-operated cafés, after baristas reported the app miscounting items, including doubling oat…
Bruce Schneier, a public policy lecturer at the Harvard Kennedy School and the University of Toronto’s Munk School, proposed a framework for deciding when to use AI by treating tasks as either 'work' or 'gym,' borrowing …
A Harvard Business School study of 150 junior consultants over 12 months found that those who treat AI as a collaborative partner—actively interrogating outputs and refining prompts—reported 40% faster skill acquisition,…
Together AI released a significant update to its inference platform, giving users complete control over performance, cost, and quality without building their own stack. The platform supports open-weight, licensed closed-…
Researchers propose UnMixNet, a physics-closed graph neural solver that embeds Maxwell-Stefan multicomponent transport PDEs, competitive adsorption ODEs, and nonlinear sensor transduction ODEs into end-to-end gas unmixin…