On Monday, Newsom signed an executive order requiring certain guidelines for AI companies doing business with California. The order calls for California to vet AI companies based on their policies ensuring that their models don’t display bias, violate civil rights or free speech, or contribute to the distribution of illegal content such as child sexual abuse material.

Just over a week ago, the Trump administration floated a framework for a federal AI policy that seeks to supersede state AI laws so as to prevent what it calls a “patchwork of conflicting state laws.” And in December, Trump even instructed the Justice Department to sue states which pass their own AI laws. Newsom’s order involves only companies that contract with the state of California, so it may not directly contravene Trump’s efforts to freeze state AI regulation. But Newsom is clearly thumbing his nose at Trump—the press release even cites the “Trump administration’s recent contracting missteps,” referring to the Pentagon’s controversial blacklisting of Anthropic.

Today’s news below.

Alexei Oreskovic
@lexnfx
alexei.oreskovic@fortune.com

Want to send thoughts or suggestions to Fortune Tech? Drop a line here.

Was this newsletter forwarded to you? Subscribe here!

Microsoft puts GPT and Claude to work together in Copilot

Microsoft has announced upgrades to its Copilot research assistant that allow multiple AI models—including OpenAI's GPT and Anthropic's Claude—to be used together within the same workflow. This includes a new Critique feature where GPT drafts a response, then Claude reviews it for completeness and citation integrity before it reaches the user. Microsoft says the workflow will eventually run bidirectionally.

The move is a direct attempt to address issues around AI errors (known as hallucinations) and boost the reliability of AI tools—problems that have slowed enterprise adoption of Copilot amid stiff competition from Google's Gemini and Anthropic's Claude. A separate Model Council feature will allow users to compare outputs from different models side-by-side.

Microsoft is also rolling out Copilot Cowork—its agentic AI tool for long-running, multi-step tasks—to members of its Frontier early-access program. The product, based on Anthropic's Claude Cowork, was first unveiled earlier this month.—Beatrice Nolan

Shedding light on Sora's mysterious death

OpenAI's abrupt decision earlier this month to kill Sora, its AI video creator, is one of the most baffling developments in the AI business—not least because OpenAI had just struck a high-profile deal with Walt Disney Co. months before.

The Wall Street Journal has new details that shed a bit more light on what happened. According to the report, OpenAI's decision to pull the plug on Sora was as big a surprise to execs at Disney as it was to the rest of us, with many Disney folks learning of the plan less than one hour before the announcement. 

But why did OpenAI kill Sora?

The direct cause was Spud, a new large language model that OpenAI is racing to put through its final training process. OpenAI has a finite amount of computing capacity on which to train its models, and Sora was simply hogging up too many resources for a product that wasn't a priority. 

As for why Sora wasn't a priority, the WSJ reports that usage peaked at around a million users shortly after its launch in September, and has declined to about half that since then. What's more, the hefty computing power necessary to create the videos means that Sora is losing about $1 million a day for OpenAI, an anonymous source told the WSJ. There's clearly still a lot more to learn about the death of Sora (and if you have any insight, please reach out: alexei.oreskovic@fortune.com), but at least we now have a chalk outline of the body and some bullet casings to examine.—AO

Mistral raises $830M for Paris data center

Mistral, one of France's top AI companies, has raised $830 million in debt to build a data center about 40 miles south of Paris. 

The AI data center is expected to begin operating in Q2 and will pack 13,800 Nvidia GPUs. Mistral will use the data center to train its latest open source large language models, according to Bloomberg. The company was founded in 2023 by former researchers from Google's DeepMind and Meta, and represents one of Europe's strongest home-grown alternatives to U.S.-based AI giants. As AI tech becomes increasingly important in the global economy and geopolitics, countries across the world are looking at creating sovereign AI and cloud infrastructure. Mistral said it's planning to build 200 megawatts of capacity across Europe by the end of 2027 "to support the demand from governments and enterprises that seek to build and control their own AI."—AO

More tech

—Nvidia’s Jensen Huang says ‘We’ve achieved AGI.’ But no one can agree on what that means. 

—Meta testing a premium subscription on Instagram. Surreptitious Story viewing.

—Uber acquires Berlin-based Chauffeur. Courting the luxury segment.

—What happens when you combine AI and 3,000 Irish bartenders? Lower Guinness prices.

—David Sacks' post-Czar role in the White House. 'More of a change in title than influence.' 

—Micron shares are down 30% since blowout earnings. Easy come, easy go.