\n<\/p>\n
Microsoft CEO Satya Nadella is the latest tech executive to offer lengthy thoughts on how AI safety might be improved.<\/p>\n
In a Saturday morning post on X<\/a>, Nadella wrote that it\u2019s time \u201cto step back and assess the trust architecture\u201d of AI.<\/p>\n \u201cWe can\u2019t treat Super Intelligence as a set of nested black boxes and simply accept or reject its recommendations, answers, and actions,\u201d Nadella wrote, using the Trump administration\u2019s preferred term for AI<\/a>.\u00a0<\/p>\n As outlined by Nadella, this approach \u201cmeans separating the model from the harness that orchestrates its work,\u201d as well as \u201cexternalizing controls and safeguards.\u201d He also called for \u201cevery meaningful model action\u201d to be documented with \u201ctamper-proof human readable evidence,\u201d and for systems where \u201can authorized person\u201d always has the ability \u201cto pause or shut down a model mid-task.\u201d<\/p>\n \u201cWe must assume a model is compromised and contain it from the start,\u201d he said. \u201cThink of it like an emergency brake.\u201d<\/p>\n Nadella\u2019s comments come as leading AI companies acknowledge more and more incidents where they seemed to lose control of their models<\/a>, and after Anthropic CEO Dario Amodei published a plan<\/a> for more cautious AI development.<\/p>\n<\/div>\n