Anthropic’s Claude Constitution: How Model Character Became a Design Discipline
The New York Times reported that Anthropic co-founder Chris Olah has spent months in confidential meetings with dozens of religious scholars, asking an unusual question for a technology company: what kind of moral entity is Claude, and should centuries of human moral thinking shape how it behaves? However you feel about the premise, the story […]
ds4: antirez’s Narrow Inference Engine for Running Frontier Models Locally
Salvatore Sanfilippo, better known as antirez and best known as the creator of Redis, has a new project: DwarfStar 4, or ds4, a small C inference engine that runs frontier open-weight models like DeepSeek V4 and GLM 5.x locally on high-memory hardware. It went up on Hacker News and picked up momentum fast, and the […]
Meta’s Muse AI assistant had a zero-day that turned it into the perfect Mac backdoor
Muse, Meta’s new AI assistant for macOS, shipped with a zero-day that let any locally running app or terminal command seize complete control of the agent: read its session token, inject its own prompts, and quietly exfiltrate whatever the assistant could reach, which given its permissions is a lot. Mac security researcher Patrick Wardle found […]
Gemini 4 Argon: Google’s new frontier model goes long
Google announced Gemini 4 Argon on September 30, and the rollout is deliberately slow. The model is currently reaching a set of trusted cyber defenders through Google’s Fairwind Program, with broader access for developers, enterprises and consumers to follow. Google says it is taking part in the US government’s voluntary process for pre-release model access […]
Jeff fine-tunes tiny models to make decisions in one forward pass
Most LLM calls in production are not conversations. They are routing decisions: which support queue does this ticket belong in, is this comment abusive, which intent did the user just express. A new open-source project called Jeff targets exactly that workload with fine-tunes of Qwen3.5 (0.8B and 2B) and Gemma 4 E2B that return a […]
80,000 organizations had AI logins stolen, and session cookies are the real problem
SOCRadar’s AI Identity Exposure Report, covered by BleepingComputer, puts a number on something security teams have been quietly dreading: infostealer logs now contain employee logins for AI platforms at serious scale. More than a million records tied to AI services span over 80,000 corporate domains. The researchers narrowed that set to 482 major enterprises and […]
Microsoft quietly retires the Copilot+ PC brand
The label is gone, the spec stays Two years after launching Copilot+ as the badge of honor for AI-ready Windows hardware, Microsoft has stopped using it. The new Surface Pro 12-inch and Surface Laptop 13-inch announced this month carry no Copilot+ branding at all, even though both comfortably clear the hardware bar the label was […]
unreal-agent brings async-first harness design to coding agents
A harness that never makes the model wait unreal-agent has been climbing GitHub’s trending list all week, at roughly 1.9k stars four days after it appeared, and the pitch behind it is simple enough to repeat in one line: the same model, doing the same work, up to 40 percent cheaper, because the harness never […]
Z.ai open-sources its whole coding agent stack with ZCode
The harness, not just the model Z.ai, the company behind the GLM model family, open-sourced ZCode this week, and the interesting part is the scope. This is not a CLI script and a README. The repository contains the desktop client, the browser workspace, the backend services, the shared UI, and the agent runtime that powers […]
Claude Opus 5.5 cuts prices and rethinks how the model writes
The price is the story Claude Opus 5.5 is out, and the most interesting number on the announcement page is not a benchmark score. Input tokens now cost $4 per million, down from $5 on Opus 5. Output fell from $25 to $20 per million. Cache reads dropped hardest, from $0.50 to $0.20 per million. […]
Xiaomi’s MiMo-V2.6 takes the open-weight crown and shows its RL receipts
A car company just published the strongest open weights on the board Xiaomi released and fully open-sourced MiMo-V2.6 on September 21, and the flagship Pro model lands at 46 on the Artificial Analysis Intelligence Index. That score puts it ahead of every other open-weight model on that leaderboard, above xAI’s Grok 4.6 and Google’s Gemini […]
Qwen Image 2.1 lands as open weights with a license catch
Alibaba’s Qwen team released Qwen Image 2.1 as open weights on September 20, and it hit the top of Hacker News within hours. The model unifies text-to-image generation and image editing in a single 7B-parameter network, adds native transparency support, and ships with day-zero integrations across Diffusers, ComfyUI, and vLLM. There is one detail you […]