engblogs

Databricks integrates Stately Cloud's expertise to enhance the reliability and scalability of its Data Intelligence Platform, empowering mission-critical AI applications across global distributed systems.

GPT-5.2-Codex ushers in a new era of agentic coding for professional software engineering and defensive cybersecurity, delivering improved long-context handling, reliable tool calls, and stronger safeguards to enable large-scale code changes, complex refactors, vulnerability research, and safer production deployment.

GPT-5.2-Codex enhances agentic coding with improved long-horizon context compaction, project-scale task performance, Windows compatibility, and robust cybersecurity safeguards.

GPT-5.2-Codex advances agentic coding and cybersecurity capabilities with improved long-context understanding, large-scale code management, and enhanced safeguards for responsible deployment.

GPT-5.2-Codex is an advanced agentic AI model optimized for complex software engineering and enhanced defensive cybersecurity, featuring improved long-horizon coding, large-scale code changes, native Windows support, and stronger safeguards for responsible deployment.

This blogpost presents a framework and 13 evaluations to quantify chain-of-thought monitorability and its scaling with test-time compute, reinforcement learning, and pretraining, highlighting when chain-of-thought monitoring can serve as a load-bearing control layer for safer AI.

The update to the Model Spec introduces U18 Principles to ensure ChatGPT provides safe, age-appropriate interactions for teens, incorporating expert guidance, enhanced safeguards, parental controls, and ongoing improvements through research and partnerships.

Comprehensive AI literacy resources empower teens and parents to use ChatGPT responsibly, understand AI limitations, and foster healthy digital habits together.

OpenAI and the U.S. Department of Energy have signed a memorandum of understanding to accelerate scientific discovery by integrating frontier AI models with national lab research and advanced computing resources.

Introduces a comprehensive framework and evaluations to systematically measure and preserve the monitorability of chain-of-thought reasoning in large language models across scaling, reinforcement learning, and increased inference compute.

A data-driven look at how Stripe Atlas startups in 2025 accelerated time-to-revenue, expanded internationally, and leaned into AI agents.

MIT CSAIL researchers demonstrate that brief alignment, or guidance, between neural networks enables traditionally untrainable networks to learn effectively by transferring internal structural knowledge rather than outputs, revealing hidden architectural biases and improving performance.