OpenAI had one of its most consequential weeks yet, marked by a major model launch, a senior leadership departure, mounting legal pressure, and a strategic retreat from one of its consumer products.
The week’s centerpiece was the launch of GPT-5.6, a new family of models comprising Sol, Terra and Luna, targeting enterprise work, coding and cybersecurity applications. CEO Sam Altman said Sol is significantly more token-efficient for coding tasks than prior versions, and the company described 5.6 as its strongest cybersecurity model yet. OpenAI cited third-party benchmarking to claim its models outperform Anthropic’s recently released Fable 5 and Opus 4.8 across coding tasks, framing the release as a direct challenge to its chief rival. Alongside the models, OpenAI introduced ChatGPT Work, a workplace-focused agent designed to handle office tasks such as drafting documents, spreadsheets and presentations. The company also reaffirmed its partnership with Microsoft, announcing that GPT-5.6 would become the “preferred model” powering Microsoft 365 Copilot across Word, Excel, PowerPoint and Cowork, a move seen as an effort to dispel speculation about strain in the companies’ relationship following reports that Microsoft had begun relying more heavily on its own in-house MAI models to cut costs.
The same week, OpenAI confirmed it is sunsetting Atlas, its ChatGPT-based web browser launched last October, redistributing its agentic browsing capabilities into a new Chrome extension and an upgraded ChatGPT desktop app instead. The shift comes amid a broader industry contest over AI-powered browsing, with competitors including Perplexity’s Comet and The Browser Company’s Dia vying for similar territory, and follows earlier internal pressure to scale back side projects such as the video tool Sora.
OpenAI also faced a significant legal setback in its long-running copyright dispute with The New York Times and The Daily News. The publishers allege that OpenAI misrepresented its ability to search training data and chat logs for copyrighted journalism, claiming a court-ordered deposition revealed the company had already conducted internal searches and built a database of tens of millions of de-identified ChatGPT conversations to assess potential infringement. The plaintiffs are asking the court to sanction OpenAI for allegedly obstructing discovery and to bar the company from relying on a chat log sample they describe as unreliable. Lead counsel Ian B. Crosby said OpenAI’s conduct suggested awareness of wrongdoing, while OpenAI spokesperson Drew Pusateri denied the allegations, characterizing the request as an intrusion on user privacy.
Compounding the week’s upheaval, Fidji Simo, OpenAI’s CEO of Applications and the company’s second-highest-ranking executive, announced she is stepping back from her full-time role due to an ongoing medical leave, transitioning instead to a part-time advisory position. Simo, who joined OpenAI’s board in 2024 and took on the applications role in 2025, had overseen the company’s business and product operations, with COO Brad Lightcap, CFO Sarah Friar and former CPO Kevin Weil reporting to her. Her departure leaves a leadership gap as OpenAI considers a possible public offering, with chief revenue officer Denise Dresser, a former Slack CEO, seen as a potential candidate for expanded responsibilities. Altman responded to the news publicly, expressing gratitude for Simo’s contributions.
The developments unfolded against a broader backdrop of regulatory uncertainty surrounding frontier AI model releases. Analysts and researchers, including Georgetown’s Mina Narayanan and Databricks co-founder Andy Konwinski, said there remains little public clarity about how models such as Sol and Anthropic’s Fable were cleared for release, with the federal government yet to finalize a formal evaluation framework. The ambiguity has fueled broader debate about industry influence over AI policy, even as OpenAI and its competitors continue rapidly deploying new, more powerful systems.
