OpenAI, Google sold AI models to blacklisted Chinese firms
What's the story
OpenAI and Google have been selling their advanced artificial intelligence (AI) models to Singapore-based subsidiaries of Chinese companies on a Pentagon blacklist. The revelation, first reported by Financial Times and later confirmed by both tech giants, highlights a major loophole in Washington's strategy to curb China's access to cutting-edge AI technology. The companies involved are Alibaba, Baidu, and Tencent, three major Chinese tech giants with parent companies listed by the US Department of Defense as potentially linked to China's military.
Policy discussion
Transactions didn't break any US laws
The transactions between OpenAI, Google, and the Chinese companies didn't break any US laws. However, they have reignited a policy debate over whether the US is focusing too much on restricting AI chips while ignoring frontier AI software with weaker export controls. This highlights a major weakness in Washington's AI strategy as commercial access to these models through cloud-based application programming interfaces (APIs) remains lightly regulated.
Regulatory gap
Concerns over 'model distillation'
The Chinese companies accessed these AI services through their overseas subsidiaries registered in Singapore, which is why the arrangements fell within existing compliance frameworks. However, security experts warn that such structures can still give sophisticated AI capabilities to organizations that Washington seeks to limit. This has raised concerns over "model distillation," a process where outputs from advanced AI systems are used to train competing models.
Access suspension
OpenAI suspended API access for some users
OpenAI recently suspended API access for some Alibaba-affiliated users after identifying concerns related to potential misuse. Google also acknowledged the ongoing threat and admitted that simple geographic sales restrictions are no longer enough to completely stop the risks of model distillation. The controversy highlights a broader shift in geopolitical thinking, with AI models increasingly seen as strategic assets rather than just software products.
Tech control
Reciprocal technology controls between US and China
The disclosure comes as both Washington and Beijing are tightening their grip on advanced AI technologies. The US has increased scrutiny of frontier AI models along with semiconductor exports, while China is reportedly planning to limit foreign access to its own advanced AI systems. This marks the beginning of an era of reciprocal technology controls between the two superpowers.
Industry disparity
Should software exports be regulated like semiconductor shipments?
The differing approaches among American AI companies also highlight the lack of a unified industry standard. While OpenAI and Google have allowed limited overseas access under compliance safeguards, Anthropic has taken a much stricter stance by blocking all Chinese entities from using its frontier models over security concerns. This episode raises a larger question for policymakers: if AI models increasingly determine economic competitiveness and national security, should governments regulate software exports with the same intensity as semiconductor shipments?