AI Briefing

The Trump administration is quietly drafting an executive order to vet AI models before public release, a stunning policy reversal triggered by Anthropic's Mythos. Cerebras filed for what would be the largest tech IPO of 2026 at a $26.6B valuation, Google accidentally shipped its proactive 'COSMO' assistant to the Play Store ahead of I/O, and SAP dropped €1B on a tiny European startup most people have never heard of, betting that tabular foundation models, not LLMs, are how AI actually plugs into business data.

01PolicyTom's Hardware ↗

White House Considers Pre-Release Vetting of AI Models, a Policy Reversal Sparked by Anthropic's Mythos

The Trump administration is drafting an executive order that would force OpenAI, Google, and Anthropic to clear new frontier models with a federal AI working group before public release, according to NYT and Bloomberg reporting. The trigger: Anthropic's Mythos model, which uncovered a 27-year-old software vulnerability in testing and was withheld from release for security reasons. This is a sharp reversal from Trump's earlier deregulatory posture and would mark the first federal pre-release gate on US AI models.

02FundingCNBC ↗

Cerebras Files for $3.5B IPO at $26.6B Valuation in the Boldest Public-Market Challenge to Nvidia Yet

Cerebras filed an amended S-1 to sell 28 million shares at $115-$125, targeting a $26.6B valuation and roughly $3.5B in proceeds, in what would be the biggest tech IPO of 2026 so far. Banks are reportedly fielding $10B in orders against the $3.5B on offer. The chipmaker, whose biggest customer is OpenAI via a $20B compute deal, is the most direct public-market shot at Nvidia's AI hardware monopoly to date.

03Product9to5Google ↗

Google Accidentally Ships 'COSMO,' a Proactive On-Device AI Agent, to the Play Store Before I/O

Google briefly published a 1.13GB experimental Android app called COSMO to the Play Store, then yanked it. COSMO is built on Gemini Nano and runs locally, with 14 pre-defined 'Skills' that fire proactively based on what the user is doing, including a Mariner browser agent, automatic calendar scheduling from chat detection, and deep research. It's effectively Google's answer to OS-level agentic assistants, and it leaked two weeks ahead of its expected I/O 2026 unveiling on May 19.

04IndustrySAP News ↗

SAP Acquires Germany's Prior Labs in €1B+ Bet That Tabular Foundation Models, Not LLMs, Will Run Enterprise AI

SAP announced a definitive agreement to acquire Prior Labs and pour €1B+ into it over four years to build 'a globally leading frontier AI lab' in Europe. Prior Labs makes TabPFN, a Tabular Foundation Model purpose-built for the kind of structured business data LLMs are notoriously bad at, like predicting payment delays, supplier risk, and customer churn. The deal, paired with SAP's separate Dremio acquisition, signals that SAP thinks the next AI battleground is the spreadsheet, not the chat window.

05PolicyThe 19th News ↗

Minnesota Becomes First US State to Outlaw AI 'Nudification' Apps, With $500K Fines Per Violation

Minnesota's HF 1606 cleared the Senate 65-0 and is now law, making it the first US state to specifically ban apps and websites that use generative AI to digitally undress photos of clothed people. Survivors can sue app owners for damages and the state attorney general can collect $500,000 per violation. There is an exemption for general-purpose tools like Photoshop, but the bill could be voided if the Trump administration follows through on plans to federally preempt state AI laws.