Scoop: Trump AI framework lacks public incident reporting guidelines
The Trump administration's new AI framework does not include a process for companies to publicly report real-world incidents caused by advanced models before they're released.
TL;DR
- The Trump administration's new AI framework lacks a process for public reporting of incidents caused by advanced AI models before their release.
- This policy gap was highlighted by the Hugging Face breach and the release of OpenAI's Astra, a model that reached its highest cybersecurity risk threshold.
- The framework's development was kept largely private, with industry feedback incorporated but public incident reporting excluded from the final version.
- Current U.S. federal law does not outline AI incident reporting procedures, and state laws do not cover incidents like the Hugging Face breach.
- The EU AI Act requires providers of powerful AI models to report serious incidents to regulators, a contrast to the U.S. approach.
- Several other questions regarding government engagement, model review, and the definition of "frontier models" remain unresolved within the framework.