Meta child-safety trial opens in Oakland; DOJ probes a16z board seats

Good morning.

Today is a big day in an Oakland courtroom, and two quieter stories deserve your time as well.

Four states open their child-safety trial against Meta, arguing that Facebook and Instagram were built to keep children scrolling. Roughly $200 billion is the states' likely scenario; Meta's own lawyers say the ceiling is $1.4 trillion.

The Justice Department has spent nearly a year studying whether two Andreessen Horowitz board seats add up to one illegal one, with consequences for how venture firms sit on boards.

And ChatGPT has taken back first place in our LLM Meter, a week after losing it.

Stay curious,

Marcus Schuler

Opening statements begin Tuesday in Oakland, where California, Colorado, Kentucky and New Jersey become the first states to put child-safety claims against Meta in front of a jury.

Lawyers for the states have told the judge that roughly $200 billion is a likely penalty scenario. Meta's own lawyers put maximum exposure at as much as $1.4 trillion, close to the company's stock-market value in August 2026.

Eight jurors will advise U.S. District Judge Yvonne Gonzalez Rogers, who rules after a trial expected to run six to eight weeks. Mark Zuckerberg and Adam Mosseri are expected to testify.

Why This Matters:

• The states are attacking product design rather than user posts, the one route that gets around Section 230's shield for platforms.
• A COPPA finding could force nationwide deletion of children's data and of any models trained on it.

The Justice Department has spent nearly a year examining whether Andreessen Horowitz built an illegal board interlock across Databricks and Fivetran.

It turns on Ben Horowitz's Databricks seat and Martin Casado's Fivetran seat. Section 8 of the Clayton Act needs no proof of competitive harm, only that its elements are met. A finding could cost a director a seat or the firm its appointment rights.

SemiAnalysis reverse-engineers PJM's capacity model and shows how small technical errors can shift billions of dollars onto electricity customers.

The authors estimate that errors in PJM's Reserve Requirement Study cost ratepayers $12 billion in the capacity auctions for delivery years 2025/26 and 2026/27. Their reconstruction finds 3.8 gigawatts of reliable capacity on PJM's grid as of August 16, 2026, equal to 56 percent of the 6.8 gigawatts PJM plans to procure in an emergency auction opening September 30, 2026.

• Nvidia agreed to guarantee up to $105 billion in lease payments for OpenAI's eight-gigawatt Ohio campus, and to put $1.5 billion into the developer building it.
• Anthropic told investors its annualized revenue run rate reached $65 billion at the end of July, up from $47 billion in May.
• ChatGPT retook first place in the Implicator LLM Meter at 88, ending Claude's one-week lead after a watermark shipped that no plan tier can switch off.
• Groq raised $350 million at a $3.5 billion valuation, down from $6.9 billion before Nvidia licensed its technology and hired much of its senior team.
• India's Delhi High Court declined to grant ANI an interim injunction against OpenAI, leaving ChatGPT running while the copyright case proceeds.
• Relay shut down and its staff joined Google's Chrome team, closing an AI automation startup into a browser organization.

Stress-test an executive update for misreadings.

A short internal update rarely fails on its facts. It fails when three audiences read three different promises into the same paragraph. Run this before you send it.

Your raw input: the draft update, the decision it announces, the facts that must stay exact, the audiences receiving it, and any conclusion you do not want readers to draw.

The prompt:

Why this works: assigning distinct readers makes the model test interpretation rather than polish wording. Requiring the triggering phrase ties every warning to evidence, and the length and fact constraints keep it from inventing context.

What to use: Claude handles long, detailed internal drafts well. ChatGPT is a good fallback for updates under one page.

Glasp MCP Connector gives Claude access to the articles, PDFs, videos and Kindle passages you saved in Glasp. It retrieves your own highlighted source material inside a chat, instead of making you copy excerpts into a prompt. The connector is included in Glasp's free plan, while Pro costs $15 per month or $12.50 per month billed annually, checked August 10, 2026.

How to use it:

1. Create a Glasp account and save highlights from the material you want Claude to search.
2. Open Glasp Settings, select MCP Connector in the left sidebar and copy https://glasp.co/api/mcp.
3. Open Claude, go to Customize > Connectors, click + and select Add custom connector.
4. Name the connector Glasp, paste the server URL and click Add.
5. Sign in to Glasp, approve read-only access and ask Claude, "What have I highlighted about pricing strategy, and where was each passage from?"

Where it falls short: the connector is read-only, so Claude cannot add, edit or delete your Glasp highlights.

Credit: Midjourney

Prompt: Alice in Wonderland style, curves and stripes, pastel mountain flower footpath, woman facing camera wearing high fashion

Uber took a stake in Zipline and will add drone delivery to Uber Eats in Dallas and Houston by the end of the year, with a stated target of one million deliveries a day by the end of 2029. (TechCrunch, August 17, 2026)

Our take: One million a day works out to 365 million a year. Zipline has completed two million deliveries in its entire corporate life, across four continents, much of it medical supplies flown to clinics that had no other way to receive them. The plan is to multiply that by roughly 180 and point it at dinner.

Nobody has explained where the airspace comes from, or the batteries, or the patience of the neighbors. What has been explained is the delivery window, five to ten minutes instead of thirty. A medical logistics operation with a genuinely remarkable safety record now has a burrito quota, and the number in the press release is the one nobody has to hit until 2029.

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai