Introducing GPTs: A New Era Begins
OpenAI's GPTs let anyone build custom versions of ChatGPT by combining instructions, uploaded knowledge files, and capabilities like web browsing, DALL·E image generation, and code interpretation—no c…
Beginner Basics
Plain-English explainers: tokens, context windows, hallucinations, fine-tuning, RAG and agents.
Get started with ChatGPT right away
Learn how AI chat tools like ChatGPT can now be used without creating an account. This article explains why the sign-up wall was a barrier, what no-sign-up access looks like including typical limitati…
Tracing where the content we encounter online comes from
A new technology initiative helps researchers recognize content generated by AI tools, alongside joining the C2PA Steering Committee to advance industry-wide standards for content provenance and authe…
Reasoned alignment: how deliberation makes language models safer
OpenAI's deliberative alignment is a new safety technique built into the o1 reasoning models. Instead of training models to memorize refusal patterns, it teaches them explicit written safety specifica…
AI-powered pig butchering operations: The mechanics of romance scams
OpenAI banned accounts linked to Cambodia-based romance and investment scam operations, known as 'pig butchering' schemes. Scammers used AI models to translate conversations and generate persuasive sc…
Welcome to the Era of Intelligence
OpenAI debuted its first TV commercial during the Super Bowl, introducing AI to a massive mainstream audience. Rather than promoting a specific product, the ad positions AI as the latest in humanity's…
Diving deeper into the sycophancy issues we overlooked
This article takes a closer look at sycophancy issues that were previously overlooked. It reviews the results of our analysis, candidly examines the missteps made along the way, and outlines the speci…
Anticipating biological threats from advanced AI
Advanced AI is transforming biology and medicine, but it also poses biosecurity risks if misused. This article explains why AI systems capable of aiding research could also enable harm, and how compan…
Partnering with 400,000 educators to define the future of AI in education
OpenAI and the American Federation of Teachers have launched a five-year initiative to train 400,000 K-12 educators across the US on AI. The program positions teachers as leaders in AI adoption, offer…
AI: The Ultimate Empowerment Tool for Everyone
A practical-minded technologist reflects on why AI matters: not for its impressiveness, but for the tangible difference it makes in everyday lives. AI has the potential to open more doors for more peo…
What ChatGPT is being optimized for
OpenAI is refining ChatGPT with user wellbeing in mind. Key changes include more supportive responses during emotionally difficult moments, gentle reminders to take breaks during long sessions, and hi…
Accessible AI Models for Everyone
A new release of open-weight AI models aims to make powerful AI accessible to everyone, not just those who can afford premium subscriptions. By publishing model weights publicly, anyone can download, …
From outright refusals to safe completions: advancing output-focused safety training
OpenAI's new safe completions method for GPT-5 replaces the traditional hard refusal approach with output-centric safety training. Instead of refusing potentially sensitive questions outright, the mod…
Meet GPT-5
GPT-5 is OpenAI's newest flagship AI model, described as a significant leap in intelligence over previous versions. It delivers state-of-the-art performance in coding, mathematics, writing, health inf…
The reasons behind hallucinations in language models
OpenAI research reveals that AI hallucinations are not random glitches but predictable outcomes of evaluation systems that reward guessing over admitting uncertainty. Like students guessing on multipl…
Adolescent protection, liberty, and personal privacy
OpenAI has built layered protections for teenage ChatGPT users, including sensitive-topic handling for crises, parental controls, age prediction, and crisis escalation. The company aims to balance tee…
Setting up parental controls
OpenAI has begun rolling out parental controls for ChatGPT along with a dedicated parent resource page, letting families shape how the AI tool works at home rather than relying on device-level restric…
Fighting the sexual exploitation and abuse of children online
This article explains how OpenAI combats child sexual exploitation and abuse (CSEA) across ChatGPT, its API, and DALL·E. It covers strict usage policies banning CSAM and grooming, automated classifier…
Chinese state-linked misuse: Monitoring and influence campaigns
OpenAI banned a set of accounts linked to China after discovering they used its AI tools for surveillance planning, targeted profiling of individuals, and research on critics. The incident highlights …
Cyber fraud rings: The machinery behind online scams
OpenAI removed accounts tied to organized cyber fraud rings that used AI to write scam scripts, impersonate people, translate messages, and manipulate victims at scale. The article explains how these …