August 10, 2026

3 Minute Read

See how the latest AI and digital advocacy tools can transform your work.

THIS WEEK'S EDITION

Washington Just Regulated The AI You Rent, But Not The AI You Own

Last week the White House finished its rulebook for reviewing powerful AI models — and exempted roughly half the industry. In today's short read, we explain the open-versus-closed fight in plain English, and why the outcome should decide where your data lives.

The Rulebook Nobody Gets To Read

On August 4, White House officials met behind closed doors with OpenAI, Anthropic, Google, Meta, NVIDIA, and Microsoft to deliver terms for a voluntary pre-release review framework for frontier AI models. Qualifying models will give the government up to a 30-day look before launch. The framework has not been published, and the benchmarks that trigger review are reportedly classified.

The scope is the real story: the review applies to closed models only. Open-weight models are exempt — including Chinese ones.

Why the carve-out? Enforcement. Once weights are published, the file is mirrored worldwide within hours. Restricting them is like banning a recipe after the cookbook shipped. Regulators can reach Anthropic. They cannot reach the Chinese labs or the American open-weight model running on your own hardware.

Rented Models vs. Owned Models

Closed models are rented. Claude, ChatGPT, and Gemini live on someone else's servers. You send a prompt, an answer comes back, and you never touch the model. You are a tenant.

Open-weight models are downloaded. Google's Gemma 4, NVIDIA's Nemotron 3, and the Chinese trio of DeepSeek, Kimi, and GLM all publish their model files publicly.

Two quick terms to be precise on: "open-weight" and "open-source" are not the same thing. Open-weight means you get the finished model; open-source means you also get the recipe — training data and method. Most open models are open-weight only. NVIDIA's Nemotron is the rare open-source model that published the recipe too.

NVIDIA Rallied The Industry

On July 24, NVIDIA published a letter titled "Open Weights and American AI Leadership" with 24 co-signers, including Microsoft, Meta, IBM, Dell, Palantir, and Hugging Face. NVIDIA CEO Jensen Huang used his first-ever post on X to share it; Microsoft CEO Satya Nadella backed it the same day. It urged Washington against "premature restrictions" on open-weight models. The lobbying effort seemed to work, as the White House has backed down for now from saber-rattling about regulating open models.

The companies that didn't sign the letter are may be the most telling part of the story. OpenAI, Google, and Anthropic were initially missing. Under backlash, OpenAI and Google signed within roughly twenty-four hours and the roster has since passed seventy. Anthropic and Amazon still have not signed. Where you stand depends on where you sit.

What Open Models Buy You

Open models have several key advantages that are causing enterprises of all types to adopt them:

Control. Run an open model on your own hardware and your prompts never leave the building. For anyone holding sensitive research, data, or strategy, control over your own data is paramount.

Cost. Open models are cheaper, but how much cheaper depends entirely on which two you compare. The strongest open models run about 40% below Anthropic's top-end Claude. The even bigger win is running an open model on your own machines. You buy the hardware once instead of paying a toll every time someone on your team asks a question, and your marginal token cost is effectively zero.

What This Means For Public Affairs

A closed model is a vendor relationship governed by terms you did not write. An open model is an asset you control. Washington will scrutinize the models you rent and, for now at least, leave alone the models you own. That makes the move to open models for your most sensitive work an easy choice.

Three Ways To Operationalize This Email

🔱 Ask your vendors this question: "Where does our data go when we use your AI tool, and is it used for training?" Get it in writing. If they cannot answer, you have learned something important.

🔱 Sort your work into two buckets. Routine content that can safely go to a closed model, and sensitive material — client strategy, private data, key decisions — these should be on an open model where you control the data.

🔱 Price one open model against your current spend. Take a single recurring workflow and run the numbers on an open-weight alternative. Your most sensitive work may also be your cheapest to bring in-house.

🔱 If you're enjoying this content, please consider forwarding this email to a colleague or friend.

🔱 If you're not already a subscriber, please sign up here to stay up to date on the latest developments in political technology.

🔱 If you have suggestions for future newsletters or any other feedback, we'd love to hear from you.