The Unpredictable Allure of Cheap Chinese AI

The recent meeting between the White House and AI companies to discuss a voluntary framework for testing frontier AI models’ cybersecurity risks has sparked concerns among Senate Democrats about the administration’s lack of clarity on AI governance. Beyond partisan squabbles, deeper issues are at play.

Research fellow Sam Bresnick from Georgetown’s Center for Security and Emerging Technology notes that American companies outsource operations to cheaper Chinese AI alternatives because of their fine-tunability and openness. “You can download the weights, and your engineers can fine-tune the model for whatever specific application you want,” he said. This flexibility is a major draw for companies looking to maximize efficiency without breaking the bank.

However, this trend raises serious questions about relying on foreign AI. Bresnick noted that Trump’s approach to AI regulation has a chilling effect on the industry here: “It plays into China’s general narrative that they are the responsible actor, providing public goods – cheap, open-weight AI models that can help countries develop and enterprises do their business better.”

For American companies, this means uncertainty. If leading US models can be throttled or pulled with little notice, global customers will delay deployment, avoid integration of US models into critical workflows, and hedge by adopting Chinese or other foreign systems instead. This is precisely what’s happening with companies like DoorDash, which plans to keep its most sensitive tasks with a US provider while off-loading routine analytics to cheaper Chinese open-weight models.

Chinese AI labs have addressed surface-level security concerns by pushing open-weight systems that can be downloaded and run locally. This proactive approach aims to counter the “black-box US Government process” that may leave American models subject to sudden shutdowns.

But China is not just unpredictable – it’s also playing a long game. Beijing has used economic coercion as a tool of foreign policy, and there are early signs that Chinese officials are debating whether to block foreign users from advanced models. As Bresnick noted, “If you’re DoorDash or Airbnb and you’ve built parts of your software stack around Chinese AI, expecting those models to continuously improve, you could wake up one day and find that China has decided you don’t get the updates anymore.”

The implications are far-reaching. If American companies continue to rely on foreign AI, they may be putting themselves at risk of falling victim to economic coercion or cyber attacks. As Bresnick warned, “Cash-strapped startups quietly plugging sensitive data into foreign models with less scrutiny” is a recipe for disaster.

To move forward, the voluntary framework discussed by the White House and AI companies needs to be coupled with clearer guidelines on AI governance, investment in domestic AI research and development, and promotion of transparency and accountability in AI decision-making. Ultimately, the allure of cheap Chinese AI may be a siren song that American companies can’t resist – but it’s one they shouldn’t follow. The stakes are too high, and the risks too great. It’s time for American policymakers to take a more proactive approach to AI governance, prioritizing security, transparency, and accountability over short-term gains.