The federal government's AI testing body has received instructions to limit the public disclosure of its findings. Sean Cairncross, the National Cyber Director, has directed the Center for AI Standards and Innovation, known as CAISI, to halt the release of public reports regarding its evaluations of advanced AI models, citing concerns related to national security.
This directive aligns with an executive order from June 2026 that aims to promote advanced AI innovation while improving security measures surrounding frontier AI models.
#What is CAISI and Who Are Its Partners?
CAISI is the successor to the former US AI Safety Institute established during the Biden administration. While its name has changed, its main goal remains unchanged: to assess advanced AI models for risks associated with cybersecurity and biosecurity before they are widely adopted.
The agency has voluntary agreements in place with leading figures in AI development. Companies like OpenAI, Anthropic, Google DeepMind, Microsoft, and xAI submit their advanced models for evaluations focused on national security.
It is essential to understand that these testing agreements are entirely voluntary. Companies are not legally obligated to submit their models for assessment, as there are no mandatory licensing requirements or pre-approval processes in place.
#What Does This Mean for Transparency?
The June executive order underscores an approach that prioritizes voluntary cybersecurity assessments of frontier models prior to their broader deployment instead of enforcing strict requirements. Importantly, this does not halt CAISI’s testing program. The voluntary agreements with AI developers remain intact, and no new mandatory requirements have been introduced. The only change lies in the relocation of evaluation results from public access to internal government communication.
#What Are the Implications for Investors?
The lack of mandatory licensing grants AI developers the flexibility to innovate without awaiting governmental approval, but it introduces unpredictability concerning the evolving regulatory landscape. Companies with significant investments in safety and collaborative research, like Anthropic, might see their competitive edge diminish if assessment results remain undisclosed. Investors should monitor whether CAISI’s agreements with major AI companies continue and if any developers choose to distance themselves from a testing process that no longer offers transparent reputation benefits.