How Microsoft discovers and mitigates evolving attacks against AI guardrails

This Microsoft blog explains risks from malicious manipulation of large language models—distinguishing 'malicious prompts' and 'poisoned content'—and introduces mitigations including Spotlighting (data marking), multiturn prompt filtering, an AI Watchdog, and research updates to address a newly discovered multiturn jailbreak technique called Crescendo; the post is research and defensive guidance rather than an incident report.