Are Copilot prompt injection flaws vulnerabilities or AI limits?

The article covers a security engineer’s disclosure of several issues in Microsoft Copilot — notably system-prompt disclosure, prompt-injection risks, and a file upload restriction bypass using base64 encoding — and the ensuing debate after Microsoft judged the reports out-of-scope per its AI bug bar; the piece highlights differing views on whether these behaviors are exploitable vulnerabilities or inherent limitations of LLM-based assistants.