China’s financial institutions draw line on OpenClaw as AI frenzy spreads

Financial sector pushes back

Chinese banks, brokers and insurers are reportedly distancing themselves from OpenClaw, an open-source generative AI model that has become a flashpoint in the country’s fast-moving AI scene. It has been reported that senior risk officers and compliance teams have warned against piloting OpenClaw in trading, credit scoring and customer-facing workflows, citing concerns over model reliability, data leakage and regulatory compliance. The move marks a rare instance where finance — traditionally quick to adopt automation — is choosing caution over the get-big-fast impulse that has gripped parts of China’s tech sector.

Regulators, risk and reputational exposure

Regulators have been more explicit about the stakes. It has been reported that banking and securities regulators want firms to limit use of unvetted large models in core systems and to document safety testing, audit trails and data lineage before any commercial deployment. Why the caution? Generative models can hallucinate, leak sensitive information, or be manipulated in ways that could amplify market abuse. For institutions that manage other people’s money, reputational and systemic risks are immediate and tangible.

Broader tech and geopolitical context

The financial sector’s pushback comes amid a broader frenzy: Chinese internet giants such as Baidu (百度) and Alibaba (阿里巴巴) are racing to commercialize proprietary AI services, while startups and open-source communities are releasing competing models. At the same time, geopolitical pressures — including Western export controls on advanced chips and AI technologies — have intensified domestic efforts to build homegrown alternatives. The result is a two-track environment: rapid innovation on one side, regulatory and institutional caution on the other.

Is this a temporary brake until governance catches up, or the start of a more restrained approach to AI in regulated industries? For now, China’s finance sector appears to be drawing a clear line: enthusiasm for new models must be matched by controls that minimize legal, operational and systemic risk.