The Trump administration’s latest attempt to assess potential cybersecurity risks from advanced AI is limited and unclear, according to reports. The framework excludes open models entirely and doesn’t define what a ‘national security risk’ entails.

Following President Trump’s June executive order, AI companies like Anthropic, OpenAI and Google were invited to briefings on the voluntary guidelines but are not being told how these will be enforced. The 30-day review period for new models is narrow, focusing only on closed-source AI with state-of-the-art capabilities.

However, the lack of definition for ‘state-of-the-art’ or ‘national security risk’ leaves much ambiguity, especially concerning smaller providers aiming to stay in good standing with the White House. AI companies are not obligated to comply but are eager for clarity on releasing models without restrictions.

This framework’s vagueness could inadvertently create a regulatory grey area, making it challenging for firms to navigate while ensuring their technologies remain secure and compliant. It's a step towards oversight but one fraught with uncertainty.