Tech

OpenAI and NORAD team up to bring new magic to “NORAD Tracks Santa”

Every December, millions of families worldwide follow along as the North American Aerospace Defense Command (NORAD) tracks Santa’s⁠(opens in a new window) journey across the globe. It is a tradition full of wonder, and this year we are excited to add a few new touches to help parents make the season feel even more magical.

What to know about a recent Mixpanel security incident

Transparency is important to us, so we want to inform you about a recent security incident at Mixpanel, a data analytics provider OpenAI used for web analytics on the frontend interface for our API product (platform.openai.com⁠(opens in a new window)).

Announcing the initial People-First AI Fund grantees

Unrestricted funding for 208 nonprofits advancing people-first work nationwide.

OpenAI to acquire Neptune

Gaining deeper insight into how frontier models learn.

Accenture and OpenAI accelerate enterprise AI success

Accenture and OpenAI are collaborating to help enterprises bring agentic AI capabilities into the core of their business and unlock new levels of growth.

Funding grants for new research into AI and mental health

Introducing a new program to award up to $2 million to support independent safety and well-being research.

OpenAI takes an ownership stake in Thrive Holdings to accelerate enterprise AI adoption

OpenAI and Thrive Holdings are partnering to accelerate enterprise AI adoption.

Introducing the Teen Safety Blueprint

A framework for building AI that protects, empowers, and creates safer experiences for teens.

Introducing IndQA

A new benchmark for evaluating AI systems on Indian culture and languages.

Detecting and reducing scheming in AI models

Together with Apollo Research, we developed evaluations for hidden misalignment (“scheming”) and found behaviors consistent with scheming in controlled tests across frontier models. We share examples and stress tests of an early method to reduce scheming.

Measuring the performance of our models on real-world tasks

We’re introducing GDPval, a new evaluation that measures model performance on economically valuable, real-world tasks across 44 occupations.

How confessions can keep language models honest

We’re sharing an early, proof-of-concept method that trains models to report when they break instructions or take unintended shortcuts.