We reran our springtime disinformation tests on six AI image generators — only one company showed real safety improvements.