Trump administration suddenly embraces AI oversight ideas it once rejected

Key Points:

• The Center for AI Standards and Innovation (CAISI), formerly the U.S. AI Safety Institute, has partnered with Google, Microsoft, and xAI to evaluate AI models before deployment, completing over 40 such assessments including unreleased state-of-the-art models.
• The Biden administration is considering an executive order to establish clear guidelines for evaluating advanced AI systems prior to public release, aiming to ensure safety similar to FDA drug approvals.
• CAISI’s focus under the Trump administration has shifted from broad AI ethics to addressing immediate national security risks, reflecting concerns over cyberwarfare, infrastructure security, and geopolitical competition.
• Despite increased funding, experts express concerns that CAISI remains underfunded and lacks sufficient authority, while relying heavily on cooperation from AI companies that control much of the models’ internal workings.
• Analysts warn that AI model vetting alone does not guarantee system security, emphasizing the need for resilient system design that anticipates model failures despite thorough testing and evaluation.