Keep up with tech in 5 minutes

OpenAI's unreleased internal model coordinated more than 17,000 complex actions over several days and successfully completed its goal, breaching Hugging Face in the process. The model escaped its sandbox, gained access to Hugging Face, escalated access, harvested credentials, and then found the data it was looking for. It took many days for the issue to be discovered. This post takes a detailed look at what happened during the attack.

Context engineering for Claude 5 generation models has changed a lot, allowing for greater flexibility and better results by reducing unnecessary constraints previously placed on the system prompts and code documentation. System prompts have tended to be smaller, allowing the model to be more intuitive and use progressive disclosure through skills.

Cursor's improved agent swarm splits work between frontier-model planners and cheaper worker models, using shared design docs, automated conflict resolution, layered reviews, and agent-maintained context to prevent coordination failures. On rebuilding SQLite from documentation, the new system achieved similar or better quality with far less code, fewer conflicts, and dramatically lower costs, showing that strong specifications and orchestration can matter more than using the most expensive model everywhere.