Despite assurances to the contrary, Microsoft’s newly-rebranded Copilot AI system keeps generating all manner of inappropriate material — and this time, it’s dreaming up anti-Semitic caricatures. Copilot’s image generator, known as Copilot Designer, has such major issues with generating harmful imagery that one of Microsoft’s lead AI engineers alerted both the Federal Trade Commission and the company’s board of directors to the so-called “vulnerability” that allows for the content’s creation. In a letter posted to his LinkedIn, Microsoft principal software engineering lead Shane Jones claimed that when he was testing OpenAI’s DALL-E 3 image generator, which undergirds Copilot Designer, he “discovered a security vulnerability” that let him “bypass some of the guardrails that are designed to prevent the generation of harmful images.” “It was an eye-opening moment,” the engineer told CNBC, “when I first realized, wow this is really not a safe model.” The system, when tested out by Tom’s Hardware, was glad to generate copyrighted Disney characters smoking, drinking, and emblazoned on handguns, as well as anti-Semitic caricatures that reinforced stereotypes about Jewish people and …