We analyzed the code published by 29 frontier AI labs. Here's the data (opsera.ai)

A recent analysis of 83 public repositories from 29 leading AI organizations, conducted using the ForgeScore eight-dimension software readiness assessment, revealed significant challenges in the trustworthiness and governance of AI-generated code. Despite the advanced capabilities and resources available to these frontier labs, no organization scored above 80 on the readiness scale, with an average score of 74.6. This suggests that while AI can accelerate coding processes, it does not inherently ensure that the resulting software is understandable, secure, or easy to maintain, which raises critical questions about the long-term viability and operational safety of AI-generated software. The findings particularly highlighted "Trust Boundaries" as the weakest dimension, with an average score of 71.1, indicating potentially serious risks associated with authentication and data management. Notably, issues like exposed API keys and unprotected inference endpoints were identified in some repositories, underscoring a troubling reality: working code does not equate to trustworthy code. The analysis emphasizes that as AI continues to speed up software production, enterprises must prioritize governance, operational clarity, and security to prevent compounding technical debt and exposure that could arise from rapid, unchecked development.