Agentic market $10.8B and climbing  ·  editor@gaasnews.com
Sections
HomeWhat is GaaS?PlatformsPricingGlossaryOpinionAboutContact
HomePolicy & RegulationExport Loophole
Policy & Regulation

OpenAI and Google sold AI to blacklisted Chinese firms through a Singapore side door

US export controls stop chips at the border but not model access through the cloud. Both companies confirmed the sales. The loophole is now Washington's problem, and maybe yours.

AJ
Andrew Jamerson
Founding Editor
Jul 11, 2026 · 4 min read
Pegdoll with leader and worker icons building a sPhoto: Pegdoll with leader and worker icons building a s. // GaaS News
TL;DR
  • A Financial Times investigation found OpenAI and Google sold AI services to Singapore-based subsidiaries of Alibaba, Baidu, and Tencent, whose parent companies sit on the Pentagon's 1260H list of firms allegedly tied to China's military. Both companies confirmed on the record.
  • The sales are legal: US export controls restrict chip shipments, not cloud-delivered model access through offshore subsidiaries.
  • OpenAI recently revoked Alibaba-linked accounts suspected of distillation and reported the activity to federal authorities; Anthropic bars Chinese-owned entities including overseas units.

The Financial Times reported on July 10 that OpenAI and Google sold AI services to Singapore-based subsidiaries of Alibaba, Baidu, and Tencent, three companies whose parents appear on the Pentagon's 1260H list of firms allegedly tied to China's military. Neither seller denied it; both confirmed the arrangements on the record, per the FT's findings as syndicated by Yahoo Finance.

Legal, by design of an older regime

Nothing described breaks the law, which is the point. US export controls were written to stop physical objects: they restrict advanced chip shipments to China but say nothing about a Chinese company's offshore subsidiary buying model access over the cloud, as The Next Web laid out. Google pointed to "strong protections" on subsidiary sales. OpenAI, notably, recently revoked Alibaba-linked accounts suspected of using its outputs to train competing models and reported the activity to federal authorities. Anthropic maintains the strictest line of the three, barring Chinese-owned entities including their overseas units.

The gap becomes a procurement question

The report has already reignited the export-control debate in Washington, and the political logic points one direction. A regime that stops a GPU at the port but waves through the same capability as an API call is not a control; it is a formality, and formalities with headlines attached tend to get fixed. The fix is what agent operators should watch. If cloud model access gets pulled inside the export perimeter, every multinational deployment pattern built on offshore subsidiaries inherits a compliance problem overnight, and model choice becomes jurisdiction choice. Beijing built its side of the wall first, with national agent rules that already pulled products off the market. Washington's side is now visibly under construction, and this story is the blueprint leaking early.

AJ

Andrew Jamerson

Founding Editor, GaaS News

Andrew Jamerson is the founding editor of GaaS News, covering the economics of the agent era. He started the publication to cover Agentic AI as a Service as a dedicated beat and edits every article on the site.

Be on the list when the beat breaks

One email when a platform ships, a round closes, or the ground shifts under the software stack.