Washington weighs an AI regulator; DeepSeek preps a Shanghai IPO

The Trump administration is developing plans for an independent body to vet frontier AI models before release — modeled on FINRA, the industry-funded self-regulator that polices Wall Street brokerages, and reporting to the Securities and Exchange Commission. Treasury Secretary Scott Bessent helped shape the proposal, now under White House review, and it echoes a public call from Google DeepMind's Demis Hassabis that OpenAI, Microsoft, and Elon Musk have all backed. Officials reportedly accelerated the work after a new Chinese model rattled AI stocks; President Trump has not yet reviewed it.

DeepSeek is in talks with investors for a new round at a pre-money valuation of roughly $71 billion — just a month after closing its first outside raise of about $7 billion — while beginning IPO preparations aimed at a mainland China listing. The Hangzhou lab says it needs the capital for gigawatt-scale data centers, in-house inference chips, and new AI agent products. A public debut would make China's highest-profile open-model lab one of the first frontier labs to test domestic capital markets.

Fireworks AI closed a $1.5 billion Series D led by Atreides Management, Index Ventures, and TCV, valuing the Nvidia-backed startup at $17.5 billion. Founded by former Meta engineers, Fireworks lets companies fine-tune and serve open models in production, and it says annualized revenue has passed $1 billion — five times a year earlier — while it now handles more than 40 trillion tokens a day. The round is another bet that cheaper, customizable open models will eat into demand for frontier APIs.

Google began rolling out Connected Apps for Search's AI Mode this week, letting US users link Instacart, Canva, and YouTube Music so the assistant can act, not just answer. Ask it to plan a barbecue and it scans your Google account and Gmail, assembles a cart, and hands it to Instacart for checkout; design and playlist requests route to Canva and YouTube Music the same way. Google frames it as the start of an "era of Search agents," though for now it teed up tasks rather than completing transactions itself.

Leaked Suno source code from a breach shows the AI music startup scraped enormous volumes of copyrighted audio to train its models — including roughly 114,000 hours from YouTube Music plus thousands more from Deezer and Genius — routed through commercial proxies to bypass anti-bot defenses. The trove also exposed contact and partial payment data for hundreds of thousands of customers, who say they were never notified. The details could strengthen the copyright suits Universal and Sony have already brought against the company.

New York City Mayor Zohran Mamdani's administration recommended that landlords and brokers disclose when rental listings use AI-generated or AI-edited images, part of a "Rental Ripoff Report" drawn from hearings where about 2,400 renters described deceptive practices. The concern is sharpest for remote renters who sign leases sight-unseen and later find the unit misrepresented. If enacted, it would be one of the first US city rules aimed specifically at AI-altered real estate marketing.

In a TechCrunch interview, Index Ventures co-founder Neil Rimer argued that the fortunes AI is minting will eventually be redistributed — "either voluntary or involuntary" — and urged tech leaders to give back before governments compel it. He points to 45 new AI billionaires worth a combined $2.9 trillion even as the Giving Pledge draws few new signers, and draws a Gilded Age parallel: Carnegie preached voluntary philanthropy, but the 1930s brought top tax rates near 79 percent anyway.