Modern "Vibe Coding" prioritizes speed, a risk amplified by AI tools that replicate insecure patterns from legacy data. This leads to classic vulnerabilities like Cross-Site Scripting and new vectors like Prompt Injection. This talk offers a two-pronged strategy to address this challenge. First, we demonstrate modifying AI prompts with granular security constraints—such as demanding strict input validation and enforcing memory-safe functions—to shift AI output from functional to secure. Second, we integrate this technique into a "Secure by Design" framework, using models like OWASP SAMM and ASVS to make security an effective, built-in partner. Attendees will learn practical prompt engineering techniques and how to integrate them into their development workflows, transforming AI coding assistants from security risks into security allies.