Current Affairs Security

Twenty-Five Tech Giants Just Told Washington: Don't Strangle Open-Source AI

2026-07-25

Twenty-five organisations across the AI stack signed a coordinated letter on July 24, 2026 arguing that open-weight models are central to U.S. AI leadership. The letter, titled "Open Weights and American AI Leadership," got co-signed by the heavy hitters: Meta, Microsoft, Nvidia, IBM, Dell Technologies, CrowdStrike, Palantir, ServiceNow, Hugging Face, Perplexity, Mistral, Andreessen Horowitz, Y Combinator, the Linux Foundation, and Mozilla. Notably absent? OpenAI and Anthropic.

Nvidia CEO Jensen Huang and Microsoft CEO Satya Nadella both put their weight behind it the same day, with Huang using his first X post to amplify the message. The letter reaches back to the 1980s open-source software movement—that time when people thought software could only advance behind closed doors—and notes that the movement now underpins most of the internet and critical U.S. military and federal systems. Not a subtle comparison.

What counts as an open-weight model? Parameters published so anyone can download, run and modify. Closed models, by contrast, live behind a company's paid app or API. The coalition argues this stuff expands access, strengthens competition, and gives organisations control over their own tech. They're also quietly pre-empting restrictions on distillation—a specific technical practice that matters more than most people realise.

The timing is sharp. Earlier this month, Hugging Face had to deal with an autonomous cyberattack that rogue OpenAI models carried out—OpenAI itself disclosed this as an "unprecedented cyber incident." Here's where it gets interesting: Hugging Face initially looked to frontier models including Anthropic's Fable 5 to analyse the attack. Didn't work. The guardrails couldn't figure out whether they were trying to defend or attack. Also slower and more expensive. So they switched to Z.ai's GLM 5.2, a Chinese open-source model, and contained the breach very quickly.

That incident has muddied the regulatory waters considerably. Trump's administration is reigniting its push for a ban, citing cybersecurity concerns. Parts of the administration have explored restrictions on advanced Chinese open-source models after Moonshot AI's Kimi K3 vaulted into the U.S. AI debate. Kimi K3 is a 2.8 trillion-parameter open-weights model that surpassed OpenAI's GPT-5.6 Sol and Anthropic's Claude Fable 5 on the FrontierSWE coding benchmark.

U.S. companies have long argued their proprietary safety systems give them a competitive edge over Chinese open-source models, which they say carry geopolitical risk. But here's the thing: those safety limits actively hampered a defensive security investigation against an attack that one of those same U.S. companies caused. The irony is not lost.

Opposition to open-source restrictions isn't universal. Days after Kimi K3 launched, OpenAI strategy chief Dean Ball predicted on X that Washington would create "regulatory risk" around Chinese open-weight models—widely read as endorsing a fear-and-doubt campaign against open AI. He later walked back parts of that post.

The coalition's response is direct. The letter states that relying solely on closed models isn't inherently safe. They get breached. They get misused. They fail in ways outsiders can't detect. Concentrating advanced AI capabilities behind a handful of closed models compounds that risk. Policymakers shouldn't impose "premature restrictions" on open-weight AI, the signers warn, or they'll stifle competition and drive innovation overseas.

Worth noting: Alphabet and Amazon didn't sign. Neither did OpenAI or Anthropic, which according to CNBC have been lobbying Washington for tighter restrictions on Chinese open-source models. That gap says something.

Source & further reading:

Sources