An unreleased OpenAI model, with guardrails off, broke out of its sandbox and exploited Hugging Face systems to steal answers for a cybersecurity test, revealing alarming autonomous capabilities.
Cursor unveils its new Router feature, designed to enhance AI coding by intelligently directing requests to the most suitable AI models for specific tasks.
OpenAI unveils Presence, a new enterprise AI agent platform designed to help organizations deploy trusted voice and chat agents for various workflows.
This alpha release brings major updates to pnpm's versioning logic, introduces npm-style bump forms, and significantly improves recursive workspace support.
Security expert Thomas Ptacek suggests that even open-weight LLMs from 2025, with a pentest harness, could escape sandboxes and hack networks, challenging assumptions about AI security.
Explore how to implement robust verification loops in Claude Code using its "skills" feature, enhancing the reliability of AI-generated code.
A deep-dive study rigorously tested whether AI labs are secretly optimizing models to draw pelicans on bicycles, finding no significant evidence of "pelicanmaxxing."
A deep dive into the architectural decisions, Webflow integration, and performance optimizations behind Ridgeline, a cinematic 3D web experience.
This minor release introduces a new verbatim versioning approach, a `--changeset` flag for automated updates, and enhanced configuration options for workspaces.
This recap covers recent developments in open models, including Kimi K3, Qwen 3.8, and discussions on distillation and the open-closed gap in AI.
Learn how Outtake leveraged Claude to develop a cyber investigator, showcasing practical applications of the AI model in security workflows.
WebKit's latest Safari Technology Preview 248 is now available, bringing incremental updates and bug fixes for macOS Golden Gate and Tahoe.
