OpenAI Is in Three Fights at Once — and Each One Could Change How You Use AI

OpenAI is having a week. The company is defending itself in federal court over alleged evidence tampering, rolling out a new AI model to millions of Microsoft Office users, and quietly retreating from a product it launched less than a year ago. Each story looks separate. Together, they reveal a company moving fast in every direction — and not always cleanly.

A coalition of news publishers, led by The New York Times, asked a federal judge this week to formally sanction OpenAI. The accusation: OpenAI withheld tools and datasets that could prove whether ChatGPT was trained on — and learned to reproduce — their copyrighted articles. Asking for sanctions is a serious legal move. It tells the court that one side isn’t playing fair with the discovery process, which is the phase of a lawsuit where both sides share evidence.

Think of it like a card game where one player is required to show their hand but keeps slipping cards off the table. The publishers aren’t just saying “we think you copied us.” They’re saying OpenAI is making it harder to find out whether they did. That’s a distinct and significant claim.

For everyday readers, this case is about who gets paid when AI learns from human work. If publishers win, the economics of AI training could shift dramatically. Companies might have to license the articles, books, and other content they use to teach their models — the same way a radio station pays to play a song. TechCrunch has a full breakdown of the sanction request.

Why this matters: This case could set the legal rules for how AI companies are allowed to build their products. The outcome will affect writers, journalists, and anyone who creates content for a living.

“News publishers accused OpenAI of withholding evidence in copyright lawsuit.”

GPT-5.6 Is Now Running Inside Microsoft Word, Excel, and PowerPoint

OpenAI released a new model this week called GPT-5.6, and it came with a notable announcement: this model is now the default AI engine inside Microsoft 365 Copilot. Copilot is Microsoft’s AI assistant that lives inside productivity apps — it can draft emails in Outlook, summarize spreadsheets in Excel, and build slide decks in PowerPoint. Hundreds of millions of people use these tools for work every day.

The timing is pointed. There has been growing speculation about whether Microsoft might start distancing itself from OpenAI in favor of building its own AI models. Naming GPT-5.6 as the “preferred model” for Copilot reads as a public signal that the partnership is still very much on. According to TechCrunch’s reporting on the announcement, OpenAI also says GPT-5.6 brings improvements in areas including security and reliability.

If your job involves Office apps, this update will likely reach you through a routine software update without any action on your part. The AI assistant you already have access to will simply get better — faster responses, fewer errors, more useful summaries.

Why this matters: This isn’t a niche product update. Microsoft 365 is the most widely used productivity software on the planet, which makes GPT-5.6 one of the most broadly deployed AI models in history almost overnight.

“GPT-5.6 is now the preferred model in Microsoft 365 Copilot.”

OpenAI’s AI Browser Is Dead — But the Idea Behind It Isn’t

Less than a year after launching it, OpenAI is shutting down Atlas, its standalone AI-powered web browser. A standalone browser is a full application you download and use instead of Chrome or Safari. OpenAI built Atlas to let its AI take actions on the web on your behalf — booking appointments, filling out forms, searching across sites — without you having to click through each step manually. That kind of AI behavior, where a model acts independently to complete multi-step tasks, is called agentic AI.

Atlas is going away, but those agentic features aren’t. OpenAI is folding them into a Chrome extension and a desktop app instead. That’s a meaningful shift in strategy: rather than asking people to switch browsers entirely, OpenAI will bring the AI capabilities to wherever users already are. As TechCrunch reported, the company’s broader browser ambitions remain intact.

For users, the practical takeaway is simple: don’t download Atlas. Watch for a Chrome extension instead. The underlying goal — AI that browses the web for you — is still coming.

Why this matters: Replacing a standalone browser with an extension is a smarter distribution strategy. Extensions reach far more people, far faster.

Also Happening in AI

Google announced this week that it will add labels to ads across Search, YouTube, and Discover when those ads were made using AI — a step toward transparency that The Verge covered in detail. Separately, OpenAI is officially expanding GPT-5.6 beyond Copilot into a broader model family, with improvements touching cybersecurity and other areas, per TechCrunch. And in an unexpected personnel shift, Fidji Simo — OpenAI’s number two executive who was leading the company’s work toward artificial general intelligence, the long-term goal of building AI that matches or exceeds human reasoning across any task — is stepping down from her full-time role after a medical leave, according to both TechCrunch and The Verge.

What to Watch

The copyright trial is the slow-burning story that will matter most over the next year. Watch for the judge’s response to the sanctions request — if the court rules against OpenAI, it could force the company to hand over training data it has never made public. That disclosure alone could reshape how we understand what these models actually learned. Separately, keep an eye on how quickly Google’s AI ad labels actually appear in the wild, and whether regulators in Europe or the U.S. start treating them as a model for broader AI disclosure rules.