Welcome to Eye on AI! In today’s edition: OpenAI announces a new nonprofit commission…Nvidia reportedly kept some China customers in the dark about new U.S. chip clampdown…Sources say OpenAI is in talks to buy Windsurf for about $3 billion…A new AI-powered undercover bot for cops.

In a blog post on Tuesday, OpenAI announced the members of a recently formed nonprofit commission that will inform the company’s future philanthropic efforts.

From the company’s perspective, the new commission was, in part, designed to signal that just because OpenAI wants to convert itself into a for-profit public benefit corporation (a B Corp.), its nonprofit organization, which will still own a good chunk of the for-profit entity’s equity, “isn’t going anywhere.”

The nonprofit commission includes American labor leader and feminist activist Delores Huerta; Monica Lozano, president of the College Futures Foundation and an Apple board member; Dr. Robert K. Ross, former President and CEO of The California Endowment; and Jack Oliver, an attorney and former co-chairman of Bono’s ONE campaign.

But the timing—and the high-profile nature of the commission—suggests this may be only partly about philanthropy. As OpenAI navigates increasing scrutiny around its efforts to escape control of its nonprofit, it requires approval from California Attorney General Rob Bonta to execute its plan. Bonta oversees charitable organizations in the state to ensure that their assets are used in accordance with their original charitable purposes.

Shoring up credibility

The move to form a nonprofit commission could help OpenAI shore up credibility that it remains committed to its nonprofit mission. This is essential at the moment, coming just a week after a group of nonprofit and labor organizations petitioned Bonta to stop OpenAI from buying itself out from the control its nonprofit currently exercises over it and converting to a B Corp. The group said that OpenAI had failed to protect its charitable assets, “allowing them to be diverted for private profit and subverting its charitable mission to advance safe artificial intelligence.”

OpenAI’s effort to pursue for-profit status, it continued, is not about furthering its mission, but to provide AI’s benefits “to a handful of corporate investors and high-level employees.”

The group’s petition echoes Elon Musk’s current lawsuit against Sam Altman and OpenAI, which claims that the company betrayed its nonprofit mission that he helped establish back when OpenAI was founded in 2015. Last week, twelve former OpenAI employees asked a federal judge for permission to weigh in on Elon Musk’s lawsuit with a detailed amicus brief that accuses OpenAI of abandoning its nonprofit roots and betraying the mission that originally attracted them to the organization.

Clock is ticking on restructuring

OpenAI is in a nonprofit pickle: It needs to show the nonprofit is still functioning as a true philanthropic arm, but the company itself has argued that it needs to remove the nonprofit’s controlling role in order to raise funds from investors. OpenAI must complete its restructuring by the end of the year to secure the full $40 billion funding round led by SoftBank, which was completed in March.

No matter what, the creation of the new commission marks the end of an era when it comes to OpenAI’s nonprofit. Once tied directly to the company’s mission to develop artificial general intelligence (AGI) safely for the benefit of humanity, the commission sounds more like a way to share OpenAI’s technology so other nonprofits can boost efficiency—perhaps with purchases of the very tools that helped make OpenAI rich in the first place?

“This philanthropic effort will invest in communities as our affiliated company grows in value,” wrote OpenAI’s chief global officer Chris Lehane in a LinkedIn post. “It will help put powerful AI tools in the hands of nonprofits across California, the U.S. and beyond. And it will unlock other philanthropic dollars by removing the need to spend them on overhead and staffing, making AI a force multiplier for research, medicine, education, and discovery—along with tracking progress and outcomes of the work itself.”

Most notably, Lehane emphasized that OpenAI’s board would use the Commission’s input to “shape the evolution of the nonprofit well before the end of 2025”—just in time for OpenAI to, presumably, become the for-profit of its dreams.

And with that, here’s the rest of the AI news.

Sharon Goldman
sharon.goldman@fortune.com
@sharongoldman

AI IN THE NEWS

Google belatedly releases a model card for Gemini 2.5 Pro, but expert calls it a ‘meager’ safety report. Last week, Fortune’s Jeremy Kahn and Bea Nolan reported that Google’s latest AI model was missing a key safety report, in apparent violation of promises made to the U.S. government and at international summits. Today, three weeks after first making 2.5 Pro available to the public as a “preview,” Google finally released some safety information about the model—a report called a model card. But Kevin Bankston, a senior advisor on AI governance at the Center for Democracy and Technology, posted on X that the model card was “meager,” clocking in at just six pages, and was far from a complete safety report. “This worrisomely suggests that Google hadn’t finished all its safety testing before releasing its most powerful model to all its paying customers and any Google users who want to preview it, and that it still hasn’t completed that testing even now,” he wrote.

Sources say Nvidia kept some China customers in the dark about new U.S. chip clampdown. Reuters reported that Nvidia did not alert some of its major Chinese customers before the U.S. government informed the company on April 9 that its China-specific H20 AI chip would now require an export license—news that caught cloud providers and even Nvidia’s own China sales team off guard. The sudden move is part of Washington’s escalating efforts to limit China’s access to advanced AI chips and maintain U.S. dominance in the sector. Nvidia had already secured $18 billion in H20 orders this year, but the new restrictions jeopardize that business and could severely impact its revenue from China, which brought in $17 billion last fiscal year. The company said it will take up to $5.5 billion in charges this quarter as a result of unsellable inventory and purchase obligations tied to the H20, causing its stock to fall 6% in after-hours trading.

OpenAI in talks to buy Windsurf for about $3 billion. According to Bloomberg, OpenAI is in talks to acquire Windsurf, the AI-assisted coding startup formerly known as Codeium, for around $3 billion. It would be its largest acquisition to date and the potential deal signals OpenAI is looking to strengthen its position in the increasingly competitive AI coding assistant space, with players like GitHub Copilot, Anthropic, Cognition AI, and Anysphere. Windsurf had recently been in discussions with investors to raise funds at the same $3 billion valuation, following a $1.25 billion valuation last year. Founded in 2021, the company has raised over $200 million from backers including General Catalyst, Kleiner Perkins, and Greenoaks.

A new AI-powered undercover bot for cops. 404 Media reported that police departments near the U.S.-Mexico border are spending hundreds of thousands of dollars on a secretive and largely unproven AI tool called Overwatch, which creates lifelike virtual personas to interact with and gather intelligence on targets ranging from suspected traffickers to political activists and college protesters, according to documents obtained by 404 Media. The technology, developed by New York-based company Massive Blue, is marketed as an “AI-powered force multiplier” for law enforcement and is designed to infiltrate online spaces and communicate with suspects via text and social media.

FORTUNE ON AI

OpenAI updated its safety framework—but no longer sees mass manipulation and disinformation as a critical risk —by Sharon Goldman and Jeremy Kahn

OpenAI debuts new ‘reasoning’ models and coding agent as it seeks to stay at the front of the AI pack —by Jeremy Kahn

The only way some manufacturing jobs will return stateside is if AI-empowered robots do them—and that’s a good thing —by Tom Biegala

Exclusive: Biotech startup Profluent says it has discovered AI ‘scaling laws’ for AI models used in protein design —by Jeremy Kahn

AI CALENDAR

April 24-28: International Conference on Learning Representations (ICLR), Singapore

May 6-7: Fortune Brainstorm AI London. Apply to attend here.

May 20-21: Google IO, Mountain View, Calif.

July 13-19: International Conference on Machine Learning (ICML), Vancouver

July 22-23: Fortune Brainstorm AI Singapore. Apply to attend here.

EYE ON AI NUMBERS

131

That’s how many state-level AI-related laws were passed in the past year alone, according to 2025 AI Index Report recently released by the Stanford Institute for Human-Centered Artificial Intelligence (HAI). Amid slow progress at the federal level, U.S. states are picking up the slack: In 2016, only one state-level AI-related law was passed, increasing to 49 by 2023. That has now more than doubled. The Index noted that while proposed AI bills at the federal level have also increased, few have passed.

OpenAI taps high-profile advisors for its nonprofit—right when it needs to shore up support for its plan to remove that nonprofit’s control of its business