Open Weight Unites the Closed Labs

OpenAI and Anthropic are fierce commercial rivals racing toward IPOs — but July 22 brought proof they can lobby as a bloc when open weights threaten the closed-model business. Axios reported the two labs are aligning to warn Washington about Chinese open-weight AI, Treasury Secretary Scott Bessent is probing those models, and Q2 disclosures show they spent a combined record $3.17 million influencing Congress — all while endorsing incompatible safety bills in Massachusetts.

What's new:

• Axios on July 22: OpenAI and Anthropic are warning policymakers about risks from powerful open-weight models, especially Chinese releases like Moonshot's Kimi K3 ahead of its July 27 weight drop.
• CNBC on July 21: Q2 federal lobbying hit $3.17 million combined — up 23% from Q1 — with Anthropic at $1.97 million (more than Nvidia's $1.25 million) and OpenAI at $1.2 million.
• Both companies reported lobbying on cybersecurity, copyright, cloud computing, and defense procurement.

The shared Washington script:

• Anthropic CEO Dario Amodei argues open-weight models are harder to keep safe because developers "lose the ability to revoke access, update safety guardrails or prevent misuse" once weights ship.
• OpenAI told Axios distillation of American models is a threat, echoing U.S. Trade Representative Jamieson Greer, who views Chinese distillation as "a form of IP theft."
• Former Biden White House tech adviser Suresh Venkatasubramanian pushed back: cutting off access to Chinese models would be a "big problem for doing research."

The capture charge:

• Trump AI adviser David Sacks warns scrutiny of open weights could amount to regulatory capture — safety rules that "entrench the largest companies by making it harder for competitors to release models."
• Axios and prior reporting tie the push to sanctions, liability rules, and procurement pressure rather than statute — moves that would consolidate share for closed labs ahead of IPO windows.

The Massachusetts knife fight:

• OpenAI on July 21 endorsed a Massachusetts Senate bill requiring safety frameworks and catastrophic risk assessments — without mandatory independent audits.
• Anthropic backs a stricter House measure requiring independent catastrophic-risk evaluations every six months and granting the state attorney general enforcement power.
• OpenAI spokesperson Liz Bourgeois called the Senate bill part of a "reverse federalism" play — nudging blue states toward similar rules to build a "de facto nationwide approach."
• Anthropic's Cesar Fernandez told POLITICO the House version is "the strongest AI safety proposal in the nation" and said his company is deliberately ratcheting state rules higher while OpenAI harmonizes them downward.

The lobbying math:

• The Magnificent Seven spent $21.25 million combined in Q2 — essentially flat from Q1.
• Eight major defense contractors cut lobbying 3.3% to $19.68 million.
• Anthropic's first-half 2026 total already exceeds its entire 2025 spend of $3.1 million, per Bloomberg Government filings cited by QZ.

What comes next:

• OpenAI CEO Sam Altman meets lawmakers and the White House next week on upcoming model releases.
• Massachusetts formal sessions end July 31 — a House-Senate conference will reconcile the rival AI provisions.
• No detailed U.S. safety standards exist yet for any model, open or closed.

Convina's view: This is not bipartisanship — it is market structure dressed as patriotism. Closed labs have every incentive to frame Moonshot's subsidized weights as a security crisis while quietly splitting on whether states can sue them for catastrophic risk. Anthropic's lobbying spend now rivals Oracle's; OpenAI's Massachusetts play would cap liability below what Anthropic claims to want. Enterprise buyers should read the alignment for what it is: incumbents taxing openness on the way to a public offering, not a principled safety consensus.