Technology

➚ MOVE FAST: A blessing. The UK’s antitrust watchdog cleared Amazon’s $4 billion investment in Anthropic, declining on Friday to pursue a merger review. The AI startup’s revenue there didn’t meet the threshold to warrant a probe.

➘ BREAK THINGS: A curse. The EU fines keep coming for Meta. The latest is an $101.5 million penalty for allegedly violating privacy rules because of improper storage of user passwords. The Facebook parent has racked up a total of $2.8 billion in fines since the EU introduced its data privacy framework in 2018.

OpenAI is raising billions, inviting more regulatory heat for Microsoft

THE SCOOP

Prospective new investors in OpenAI, such as Abu Dhabi’s MGX, have privately made clear that investing billions in the startup would require taking an ownership stake in a for-profit version of the company — a move that would also give shares to its largest investor so far, Microsoft, according to people briefed on details of the ongoing negotiations.

The shift will immediately relieve investors, who are looking to put in $6.5 billion in the latest fundraising round, of risks that come with oversight by nonprofit board members, who have little involvement or stake in the day-to-day business of OpenAI.

But the biggest consequences for Microsoft may come from governments. Owning a part of OpenAI could attract more scrutiny from US and European regulators, who are already looking into whether their tie-up is anticompetitive.

Microsoft has been defending itself by arguing that it does not actually own a piece of OpenAI. Instead, it has an agreement to receive profits that are capped, once the company hits a specific multiple of its investments in the company, which total around $13 billion.

OpenAI CEO Sam Altman was in New York this week for the United Nations General Assembly, which he used to urge world leaders to accept the structural change and Microsoft’s ownership stake in OpenAI.

Altman’s pitch to European officials is that his firm can’t raise money without updating its corporate structure, and it needs those funds to continue to improve its technology. And he’s been arguing that OpenAI’s technology is the answer to Europe’s vexing decline in productivity, since its models can be fine-tuned for the region and on local languages, and run on data centers on European soil.

One idea floated in the negotiations with investors is to end the deal that makes Microsoft OpenAI’s exclusive cloud provider, allowing the startup to also use rivals, according to a person familiar with the matter. That could be looked upon favorably by competition regulators in Europe. Ultimately, Microsoft would have to agree to these terms and would likely be compensated for the concession.

Ending the exclusive arrangement might not dampen Microsoft’s fortunes, though. There is a growing realization that the data center capacity needed to train and run future generations of AI models is so large that no single company can provide all of it. And in some ways, Microsoft holds the cards in OpenAI’s fundraising negotiations because it’s first in line to receive OpenAI profits.

Microsoft, OpenAI and MGX declined to comment.

The size of a corporate loan that TikTok’s Chinese parent, ByteDance, is poised to sign, according to Bloomberg. It would be the biggest dollar-denominated corporate loan in Asia, outside of Japan, and comes as TikTok fights an effort to ban it in the US. Meanwhile, a Chinese stimulus plan announced this week helped buoy the shares of Tencent and Alibaba, which has been hit by consumer gloom amid a sluggish economy.

ChatGPT polling help. The AI Policy Institute turned to one of its most vocal critics — and ChatGPT — to change the way it’s been polling residents about California’s controversial AI bill, SB 1047 after being accused of bias in its surveys.

“I had criticized them in public, on Twitter, on podcasts,” Dean Ball, a research fellow at George Mason University’s Mercatus Center and one of AIPI’s most vocal critic, said in an interview. When AIPI’s executive director Daniel Colson called up Ball for help, his attitude changed. “I was like, oh, okay, these guys actually are interested in finding the truth.”

Ball and Colson uploaded the entire text of the bill, along with some critical analysis, to ChatGPT and let it summarize the legislation. Then Ball wrote an argument against the bill and Colson wrote one supporting it.

The new, neutral poll didn’t change the top-line result. Most Califronians still support the bill. Ball still believes voters aren’t the best people to determine whether it’s a good idea, but he learned from some of the other questions in the poll, like how voters feel about the concentration of power in AI.

California Gov. Gavin Newsom will decide in the next few days whether to sign or veto the bill.

Trustworthy AI. In March, everyone was talking about “AI safety,” a catchall phrase that was in danger of losing its meaning. So it’s not surprising that Microsoft has coined a new phrase to convey that its AI products are safe, reliable and secure: Trustworthy.

In a blog post titled, “Microsoft Trustworthy AI: Unlocking human potential starts with trust,” the company announced a slew of new product updates that use the secure hardware enclave inside Nvidia chips. That sounds fancy but it’s basically like a little computer inside the chip that’s kind of like those safes inside hotel room closets. It’s meant to hold just the valuable stuff. Your Macbook has one, too.

Earlier this week, I spoke to Sarah Bird, Microsoft’s chief product officer for responsible AI (now that I type that, I realize her title may need to change) about the company’s continuing efforts to keep AI in line.

“The breadth and the depth of what we need to do is so much more than we anticipated when we started this work,” she told me.

Secure enclave essentially allows AI models to run on servers while fully encrypted. That wasn’t possible before, and it’s still not possible to run the most powerful AI models that way, Bird says, but Microsoft is working on it. Bird’s honesty about the challenges rings true. It’s taken Microsoft longer than people thought to roll out its AI products and it’s been cautious about adding too much capability too fast.

• Latvian president: Russia won’t attack NATO ‘in a classic military way.’
• Saudi Arabia to expand cloud seeding program.
• YOLO Mitch? McConnell speaks out — on Trump, tariffs and beyond.