In partnership with
Welcome, AI enthusiasts
The Justice Department has taken OpenAI's side in the copyright case brought by the New York Times, telling a Manhattan judge that training AI on news articles is protected by fair use. It is the first time the government has entered any of these lawsuits, and it arrived days before both sides were due to file for summary judgment. Let's dive in!
In today’s insights:
Trump sides with OpenAI in NYT Copyright Case
Meta's New AI Beats China's Best Models
Anthropic Open Sources Its Commerce Agents
Read time: 4 minutes
LATEST DEVELOPMENTS
NATIONAL SECURITY
⚖️ Trump sides with OpenAI in NYT Copyright Case
Source: Getty Images
Evolving AI: The Justice Department backed OpenAI as a national security interest in the NYT copyright suit.
Key Points:
Government lawyers filed in the New York Times case, the first time the US has entered an AI training lawsuit.
The filing said the country would fall behind rivals abroad if the court ruled against OpenAI.
It warned that only the richest tech firms could afford to license training text and keep building models.
Details:
Trump's Justice Department filed the statement of interest late Tuesday in a Manhattan federal court, days before both sides were due to file for summary judgment. The filing says AI models help national security officials draw inferences from the unprotected facts inside news articles. Stanley Woodward and two other senior government lawyers told the court that the publishers suing OpenAI are reading fair use too narrowly, and that training clears the transformative bar courts apply. The filing does not bind the judge, and the department never saw the evidence gathered in the case.
Why It Matters:
Anthropic was cut off from government work in February on national security grounds, and a federal judge struck that designation down in August. Washington now uses the same phrase to argue OpenAI should keep taking published work without paying for it. Every article and post you have written online is training data, and this case decides whether anyone owes you for it.
Talk to your AI tools the way you'd talk to a colleague.
You don't send a colleague a three-word brief. You explain the context, the constraints, what you've already tried. But typing all that into ChatGPT takes forever — so you don't.
Wispr Flow lets you speak your prompts instead. Talk through your thinking naturally and get clean, paste-ready text. No filler words. No cleanup. Just detailed prompts that actually get you useful answers on the first try.
Millions of users worldwide. Works system-wide on Mac, Windows, and iPhone.
MODEL PRICING
🐉 Meta's New AI Beats China's Best Models
Evolving AI: Muse Spark 1.3 scores above Kimi K3 and costs a third as much to run.
Key Points:
Muse Spark 1.3 now ranks ahead of every model any Chinese lab has released.
every model scoring 59 or higher costs more per task to run than Muse Spark 1.3 does.
Zuckerberg teased a larger model codenamed Watermelon and open weights for the Muse Spark line, without attaching a date to either on.
Details:
Mark Zuckerberg announced Muse Spark 1.3 on Wednesday and kept token prices unchanged from version 1.2, at $1.25 per million input and $4.25 per million output. Muse Spark 1.3 is Meta's fourth model in this line since April, and three of the four have arrived since the middle of July. The model generates 235 output tokens per second and finishes coding jobs using about 20% fewer tool calls and 25% fewer tokens than Muse Spark 1.2 did. Meta is serving it through Muse Code and the Meta Model API with a one million token context window.
Why It Matters:
Kimi K3 sold work the American labs charged far more for, then started scoring alongside them, and Washington answered by weighing bans on Chinese models. Meta now sells a model that scores higher than Kimi K3 and costs less to run, so an American lab is again the cheapest place to buy work this good. You can choose on price and still hand the job to a model which most American corporation can adopt.
OPEN SOURCE
🔓 Anthropic Open Sources Its Commerce Agents
Evolving AI: Claude runs inside two working agents that any store can now clone and rebuild.
Key Points:
Retailers already running shopping agents on Claude report carts up to 35% larger and shoppers 60% more likely to finish a purchase.
The repository ships under an Apache 2.0 license with runnable versions for retail, travel, telecom and ticketing platforms.
Shopify and Priceline already run commerce agents on Claude, with Visa and Mastercard working alongside Anthropic on the release.
Details:
Claude Commerce Agents splits the job across two separate agents standing on opposite sides of the counter. The shopping agent lives inside a store's app and searches the catalog, compares options, remembers what a customer said months ago, builds the cart and answers questions about orders and refunds in the same conversation. The merchant agent reads a store's own sales history, flags stock about to run out before a promotion and drafts campaigns to move it, with a person approving every change before it goes live. The guardrails run inside the tool call, so the backend the shopping agent reaches has no method for charging a card.
Why It Matters:
Anthropic decided how hard these agents are allowed to push a sale, and it wrote that decision into code any store can download today. The teams copying it will swap in their own catalog and their own brand name, and most will never open the file holding those limits. The agent you talk to in a store next year will follow rules that no shopper and no regulator had a say in.
👀 Click on the image you think is real
QUICK HITS
🔷 Google shipped Gemini 3.8 Flash today, its third Flash release in six weeks, now scoring within a point of Claude Opus 5 on agentic coding at Flash pricing.
🎖️ The Army awarded Palantir $127M to build eight AI-powered TITAN battlefield systems, moving the program from prototype to production.
🥽 Norway's digitalisation minister said the government will regulate smart glasses more strictly and may ban facial recognition in public.
🦾 Robotics startup Generalist hit a $3B valuation, up from $2B in June, after a $200M extension led by 8VC.
📈 Trending AI Tools
*partner link