Generate a 500-word article outlining a risk assessment matrix for AI agent prompts.  The matrix should categorize potential risks associated with poorly designed or malicious prompts, considering factors such as data privacy violations, biased outputs, unintended consequences, and cost inefficiencies.  For each risk category, detail potential mitigation strategies.  The article should include at least three real-world examples of prompt failures and their consequences, and conclude with best practices for designing safer and more effective AI agent prompts.  The final output should be formatted as a structured article with clear headings and subheadings, using bullet points where appropriate for easy readability.  The article must also propose a scoring system for risk assessment based on likelihood and impact of the potential risk.