US Preps AI Security Order That Omits Mandatory Tests

Summary

Bloomberg's Michael Shepherd discusses a forthcoming Trump executive order on AI security that avoids mandatory pre-release government reviews of AI models. Instead, the administration plans to expand existing cybersecurity information-sharing programs to include AI companies. The order reflects a tension between promoting innovation and addressing security concerns, highlighted by recent releases like Anthropic's mythos model.

• The Trump administration is preparing an AI security executive order without mandatory pre-release tests.
• The order will augment existing cybersecurity information-sharing programs for AI companies.
• The White House has discouraged Anthropic from wider release of its mythos model over security concerns.
• The NSA has been exploring use of mythos to find gaps in Microsoft products.
• White House chief of staff Susie Wiles is taking a direct role in AI policy.
• The order could pave the way for wider federal access to AI models like mythos.
• There is pressure on the administration to respond to AI security concerns without over-regulating.