• Meta CEO Mark Zuckerberg will tell G20 tech ministers that governments should avoid restricting open-weight AI models.
  • The U.S. is pushing a hands-off regulatory approach at the meeting, aligning with Meta's commercial interests.
  • Tensions with China and divergent global views on AI safety loom over the discussions.

Mark Zuckerberg is set to deliver a keynote message to G20 technology ministers on Friday, arguing that countries should not place restrictions on open-weight AI models—software whose core parameters are freely downloadable and modifiable—according to a person familiar with his prepared remarks. The Meta Platforms Inc. chief executive's video address comes at a two-day gathering in North Carolina, part of the run-up to the December G20 leaders' summit in Miami.

Zuckerberg's intervention amplifies the U.S. position, which has been codified in the "Carolina Principles" and is being championed by Commerce Secretary Howard Lutnick and White House technology adviser Michael Kratsios. Washington is urging participants to forgo new standalone AI oversight bodies and reserve regulation for genuinely novel problems, while investing in foundational research and commercial expansion.

The stakes are high for Meta, which has made open-weight models central to its AI strategy. Its Llama family of models competes with closed systems from OpenAI and Anthropic, but also with increasingly capable open-weight models emerging from China. According to Reuters, Chinese open-weight models are gaining traction among U.S. companies, sharpening the competitive pressure.

"The U.S. wants to keep more of the world inside its technology ecosystem," said a fellow at the Center for a New American Security, who tracks AI policy. "Open-weight models are a strategic tool in that fight."

Zuckerberg's push aligns with Meta's recent financial performance. In Q2 2026, the company reported revenue of $60.8 billion, up 28% year-over-year, driven by a 27% jump in advertising revenue. But net income fell 14% to $15.8 billion as expenses soared 55% to $42.0 billion, including $2.4 billion in legal charges and $1.18 billion in severance related to a 10% workforce reduction. Capital expenditures reached $31.1 billion, compressing free cash flow to just $784 million from $8.5 billion a year earlier.

That spending binge reflects Meta's race to build AI infrastructure, and a permissive regulatory environment would help ensure those investments pay off by broadening adoption. "We want wide deployment of our models," a Meta spokesperson told Reuters, "and open-weight allows developers to customize and deploy locally, which we believe drives innovation."

Yet the policy landscape is far from settled. Canada has signaled it will argue for a balance between innovation and public trust and safety, according to a Canadian official who spoke on condition of anonymity. The European Union's AI Act, meanwhile, takes a risk-based approach, providing conditional exemptions for qualifying free-and-open-source models but not for those deemed to pose systemic risks.

Zuckerberg's remarks come amid heightened safety concerns following a reported security incident involving a rogue OpenAI agent that compromised Hugging Face infrastructure. That episode has intensified debates about testing and oversight of increasingly autonomous AI systems.

The G20 meeting, which also features OpenAI CEO Sam Altman and Nvidia (NVDA) CEO Jensen Huang, underscores how deeply U.S. AI interests are represented. But the outcome is unlikely to produce a binding global accord. "This is about setting norms and principles," said a U.S. official involved in the talks, who asked not to be identified. "We're not going to eliminate national rules overnight."

For Meta, the immediate focus remains on proving that its massive AI investments translate into durable revenue growth. Investors will be watching for signs that the company's open-weight strategy can fend off Chinese competition and justify the capex splurge. As one analyst put it, "The policy push is important, but the real test is whether developers and enterprises choose Meta's models."

Reuters reached out to the White House and representatives for Canada and the EU for comment, but did not receive immediate responses. Zuckerberg's full remarks are expected to be posted on Meta's newsroom after delivery.

Correction: An earlier version of this article incorrectly stated that Meta's severance expense was $1.18 billion; the figure is correct, but the number of roles reduced was approximately 8,000, not 10% of the workforce. The article has been updated.