On July 24, Jensen Huang used his first-ever X post to put a letter in front of Washington: “Open Weights and American AI Leadership.” Satya Nadella and Sundar Pichai backed it publicly that same day. Within five days, the signatory list went from 25 to 77 companies.

Two big names are missing from both versions: Anthropic and Amazon.

For marketing, the question isn’t open versus closed. It’s who’s accountable

Here’s the part almost no coverage connected back to a marketing team’s actual day-to-day: the open-weights fight playing out in Washington is, underneath, the exact same question any brand should ask before wiring an AI agent into its campaigns, its content, or its customer data. It’s not “is this model cheaper or more open?” It’s “who’s accountable when this goes wrong?”

An open-weight model you run on your own infrastructure gives you full control and, usually, a lower price tag. It also gives you zero support the night an agent generates something that violates a client’s data policy or publishes something that damages the brand on a weekend at 2am. A closed model from a lab like Anthropic costs more, but comes with a service agreement, an audit trail, and an actual company you can call. That distinction never shows up in a headline about “national security” or “AI leadership,” but it’s exactly the one a marketing director needs to settle before automating anything that touches brand voice or customer data.

What the letter actually says, and who’s on it

Per Benzinga, the letter’s core argument is that America’s AI future shouldn’t depend exclusively on a handful of closed frontier systems, and that policymakers should preserve an ecosystem where organizations can download, adapt, inspect, and run advanced models on infrastructure they control.

Pichai put it plainly: “Very happy to support this on behalf of Google.” Nadella called open-weight models “essential to a healthy AI ecosystem,” strengthening American competitiveness while protecting national security. DeepMind’s Demis Hassabis added: “a strong and secure open ecosystem is important for the world to benefit from AI.”

The signatory list grew fast. Meta, OpenAI, Cisco, Cloudflare, GitHub, Hugging Face, and dozens more joined between July 24 and July 29, according to Forbes. Neither Anthropic nor Amazon has given an official explanation for their absence, and that Forbes piece says it exactly right: “neither has explained the absence, and it would be a mistake to assume the reason.”

Anthropic’s actual position, more nuanced than what circulated

This is where most of the coverage oversimplified. The version that spread was that Anthropic opposes open weights to protect its own closed-model business. Dario Amodei answered that accusation directly, and per Techdirt, his actual position is narrower than a flat “no.”

Amodei was explicit: “we don’t think anyone should ban open weight models… but we do think the US should ban all the conditions that make quality open weight models possible.” That’s a subtle but real distinction. He isn’t asking to censor the output, he’s asking to control the inputs, specifically chip exports to authoritarian governments and industrial-scale distillation, the far cheaper compute process some labs use to copy a closed model instead of training one from scratch.

Amodei also challenged the core premise of the letter he didn’t sign: “I don’t agree with the letter’s assertions that open-weights models necessarily make it easier to develop safeguards or that broad access to capabilities necessarily helps defenders more than attackers.” That’s a specific technical disagreement, not a blanket anti-openness stance. He added a third, harder policy ask on top: mandatory pre-release safety testing for any sufficiently capable model, open-weight or closed.

What this means for a marketing team running AI in production

We’ve already written about how much marketing spends on AI tokens without measuring it, and about why your AI marketing needs a verifier, not a smarter model. This open-weights fight is the same governance question, one level up: it doesn’t matter how well an agent writes or how cheaply it runs if nobody can audit what it did and nobody is accountable when it fails.

If your team is evaluating AI agents to generate content, run campaigns, or touch customer data, “open versus closed” is the wrong question to start with. The right one is: what record exists of every decision the agent made? Who is contractually responsible if that content violates a data policy or damages the brand? Is there a real service agreement, or are you running a model you downloaded that nobody else will support when something breaks on a Friday night?

We’ve already seen Anthropic embed itself directly inside HubSpot and Canva, precisely because martech tools need that accountability chain, not just model capability. At IQ Source, our marketing automation work runs on models from labs you can audit and hold to a real service agreement, exactly because the question that actually protects a brand was never how open the model is. It’s who answers when something goes wrong with your company’s name attached to it.

Automate your marketing with a vendor you can actually audit