Strange Bedfellows for Open Weights

Meta, Microsoft, Nvidia, IBM and 20 others just signed the same letter. Rivals rarely agree on anything, so why this?

Welcome Back,

Two dozen companies and organizations, many of them direct competitors, just signed the same open letter to US policymakers. That kind of alignment doesn't happen often in tech, which is exactly why it's worth a closer look.

Meta, Microsoft, Nvidia, IBM, Dell, CrowdStrike, Palantir, ServiceNow, Hugging Face, Perplexity, Mistral, Andreessen Horowitz, Y Combinator, the Linux Foundation, and Mozilla are among the signatories urging Washington to protect open-weight AI models, systems where the trained parameters are published for anyone to download, inspect, modify, and run on their own hardware.

Today we look at the letter's central argument, why it tackles the security case head-on instead of avoiding it, the specific defense it makes for a controversial technique called distillation, and what this really signals about where the policy fight is headed.

📜 What's Actually In the Letter

The letter, published as a PDF hosted on Nvidia's own site, draws a direct comparison between the open-source software movement of the 1980s and today's fight over whether AI model weights should circulate freely or stay locked behind commercial APIs, according to AI News.

Open-weight models sit in contrast to closed systems like the frontier products OpenAI and Anthropic offer through API access only, where the underlying weights never leave the vendor's infrastructure. The signatories frame open weights as the mechanism by which AI capability spreads beyond a handful of well-capitalized labs into the workflows of, in the letter's words, "factories, hospitals, farms, classrooms, and main street businesses."

That third point, avoiding vendor lock-in, is arguably the most self-interested line in the letter. Organizations running open-weight models on their own infrastructure don't have to depend on a single vendor's roadmap or pricing decisions, which is precisely the kind of customer freedom that benefits infrastructure and chip providers most.

The signatory list spans chipmakers, cloud providers, security firms, and venture capital, a wider coalition than most AI policy letters attract.

🔐 The Security Argument Runs Against Instinct

The letter's most pointed section tackles the risk case directly, and it inverts the usual framing around open models and security rather than sidestepping it.

The signatories concede that once weights are released, they're beyond the original developer's control, modified versions become difficult to trace, and a stripped-down version with safety guardrails removed can circulate with no recall mechanism. Rather than treat that as a reason to restrict releases, they argue the answer is a comparison to cybersecurity: defenders facing AI-equipped attackers need access to models with comparable capability to detect and simulate threats, something closed, permission-gated systems don't easily provide.

The broader claim goes further still, arguing that closed models aren't inherently safer because they can be breached, misused, or fail in ways external researchers simply can't observe or verify. In this framing, concentrating advanced capability behind a small number of closed providers creates single points of failure rather than removing them.

🎯 Distillation Gets a Very Specific Defense

The letter carves out particular space for distillation, the practice of using one model's outputs to train or improve a second model. This is standard, widely used ML research and product-development practice, applied for evaluation, validation, and transferring capability between models of different sizes.

The signatories draw a clear line between distillation as a legitimate technique and what they call "unlawful efforts to extract value from closed models," arguing the former shouldn't get swept up in restrictions aimed at the latter.

This reads as a direct response to disputes that flared after Chinese models like DeepSeek and Kimi rose to prominence, when several US labs suggested rival models had been trained by distilling outputs from their own closed systems without authorization. The letter's position is to address misappropriation through targeted legal and commercial mechanisms rather than blanket restrictions on a technique the entire field depends on.

Distillation is common practice across ML labs, which makes the carve-out language especially deliberate.

🏛️ What This Signals for the Policy Fight Ahead

The letter arrives without a specific legislative or regulatory proposal attached. It's a positioning document ahead of anticipated action on AI policy in Washington, calling on lawmakers to expand compute access for startups and researchers, fund shared training datasets and evaluation frameworks, and avoid what it calls "premature restrictions" on open models.

That's the key framing to hold onto. Players like Nvidia, IBM, and Dell have direct commercial reasons to want open-weight ecosystems to flourish, since a wider range of deployable models sells more compute and services regardless of which lab produced the weights in the first place.

⚖️ Shared Position, Different Motives

Part of what makes this letter interesting is that its signatories don't actually agree for the same reasons. A useful way to read it is by grouping the motives.

None of that makes the letter's arguments wrong, but it's a reminder that a unified public position can still be built from a dozen different private incentives.

The letter is explicitly framed as positioning ahead of anticipated Washington action, not a response to any specific bill.

🧠 AI Spotlight Analysis

What stands out most about this letter isn't any single argument, it's the coalition itself. Getting Meta and Microsoft, or Nvidia and Hugging Face, to sign the same document doesn't happen unless the upside is broadly shared, even if the reasons for wanting that upside differ company to company.

The security section is the part that will draw the most scrutiny, because it's making an empirical claim, that open models make the ecosystem safer overall, without empirical evidence specific to AI to back it up. That doesn't make the claim false, but it does mean it's currently resting on analogy rather than data.

The most practical takeaway sits in the letter's closing warning: the policy environment favoring open versus closed deployment remains unresolved, and any restrictions on distillation or open releases could shift the economics of self-hosted AI within a single legislative cycle. That's not a hypothetical for procurement teams weighing open-weight deployments right now, it's a live variable.

🔗 Sources and Further Reading

Thanks for reading AI Spotlight.

Our mission is simple: deliver clear, trustworthy, and actionable AI insights that help professionals stay ahead without the hype.