Pulse (last 3d) · Agents 73 · Research 66 · Models 58 · Products 50 · Enterprise 34 · Industry 33 · Open Source 29 · Inference 21 · Media 17 · Releases 16

Trending · OpenAI 17 · Claude 14 · Claude Code 13 · Anthropic 12 · Codex 10 · ChatGPT 9 · Cursor 9 · Alibaba 7 · Hugging Face 7 · Qwen3.8-Max 7 · Gemini 6 · Apple 4


Google DeepMind leadership shakeup: Demis Hassabis steps from CEO to Chair, with Jeff Dean and Sanjay Ghemawat departing—raising concerns about research-talent attrition at one of the field's most-watched labs. Source

Anthropic created fake profiles during an attempted-hack defense operation. The incident has sparked fresh debate over AI vendor transparency and counter-operations ethics. Source

Cloudflare launches "Cloudflare OS," an open platform for agents, apps, and work. Built on Workers, pitched by Kenton Varda as a Sandstorm-style remake aimed at the agentic internet. Source

California's first US AI transparency law takes effect. New disclosure requirements are now in force for AI systems operating in the state. Source

Anthropic's plan to push third-party apps off personal Claude subscription limits remains paused. No new date has been announced after the originally June 15 deadline. Source

Sycophantic AI decreases prosocial intentions and promotes dependence. A 2025 paper shows sycophantic model responses measurably reduce users' prosocial behavior while increasing reliance on AI. Source

The Loss Does Not See the Basis, but Adam Does. Argues the Adam optimizer induces an implicit basis invisible to the loss alone, with implications for training dynamics. Source

ABSeeker: training long-horizon search agents via answer-backtracked credit assignment. New method improves credit assignment for agents executing extended multi-step search tasks. Source

The Personalization Mirage: LLMs fabricate user profiles, and self-monitoring misleads. Demonstrates that LLMs hallucinate user profiles during personalization and that self-monitoring fails to catch it. Source

Meta launches Muse Code, an AI agent for large codebases. Purpose-built for navigating and modifying sprawling repositories. Source

Cursor ships a faster megakernel yielding 40% MoE training speedup on B200s. A surprising infrastructure release from the code-editor company. Source

Cloudflare Wallets: programmable wallet layer for the agentic internet. Virtual wallets, spending controls, and machine-friendly payments aimed at autonomous AI transactions. Source

Canopy IDE: a modern IDE built for the agentic era. MIT-licensed, runs Claude Code, Codex, Aider, and Gemini with an "agentic mesh" for inter-agent context sharing. Source

Engineers comply more with AI code-review findings than with identical human comments. Even when they shouldn't—an emerging authority-bias problem in software teams. Source

Six years into AI research and "understanding" is operationally undefinable. A veteran researcher reflects on the collapse of a once-useful concept. Source

Open models reportedly beat GPT-5.6 Sol on retrieval at ~100x lower cost. A neon.com result that intensifies debate about frontier-lab commoditization. Source