This paper evaluates the compliance of major AI companies with voluntary commitments based on voluntary AI governance guidelines announced by the White House, the G7, Blechley Park, and Seoul. Based on the White House's eight voluntary commitments for 2023, we developed a detailed evaluation scale to score companies' publicly disclosed actions. The results revealed significant differences in compliance across companies. OpenAI, which received the highest score, achieved 83%, while the average score was only 53%. In particular, compliance with model weight security commitments was extremely poor, with an average of 17%, with 11 out of 16 companies achieving 0% compliance. This highlights a clear structural deficiency that must be addressed in future AI governance initiatives: proactively disclosing how companies implement their public commitments to ensure accountability and ensuring the verifiability of such disclosures. Finally, we present three policy recommendations to address issues of unclear commitments, the role of complex AI supply chains, and public transparency.
