White House AI Testing: Open Models Face New Scrutiny

The White House is reportedly preparing to include powerful open-weight AI models in its voluntary pre-release cybersecurity testing framework. This could narrow a gap for businesses choosing artificial intelligence vendors. This reported change would mean open models could be included once they reach advanced capabilities. This softens an earlier position where officials said the voluntary review would not apply to U.S. open-weight models. Here's the thing: Open-weight models are becoming more attractive to enterprises like banks and retailers. They can be downloaded, customized, and hosted privately, offering more control over cost, data, and model behavior. But this control also creates a different risk profile. Once released, a powerful open model is difficult to recall, and its safeguards can be modified or removed. The White House's executive order set up a voluntary framework for developers to provide early access for cybersecurity evaluation. This was intended to strengthen cybersecurity without mandatory licensing. What's interesting is that this balance becomes more complicated as open models grow more capable. For financial institutions, this is a practical matter. A bank might prefer an open model for internal control, but if that model can assist with cyber exploitation or be modified to weaken safeguards, the bank.