Microsoft published a provisional Code of Conduct for its MAI models, setting limits on weapons assistance, dangerous-substance procurement, deceptive behavior, autonomous goals, and concealed
Microsoft published a draft Code of Conduct for its in-development MAI models, opening it to six weeks of public consultation. The rules prohibit assistance with weapons, dangerous-substance procurement, cyberattacks and harmful manipulation. They also require human control: models must not pursue independent goals, evade shutdown or hide actions. Microsoft says MAI models “will not tamper with chain of thoughts or code” or communicate in “neuralese.