Securing generative AI: 5 action items to protect your organization

The report summarizes current threats to GenAI/LLM applications—including sensitive data leakage, guardrail evasion, jailbreak techniques, and risks from proliferating smaller/local models—and provides defensive recommendations. Key findings include that 90% of successful attacks result in data leakage, 20% of jailbreak attempts bypass guardrails, and adversaries often need only ~42 seconds and ~5 interactions to succeed; recommended mitigations include strict data governance and anonymization, context-aware guardrails with adversarial testing, prompt validation and classifier LLMs, throttling/honeypots, behavioral monitoring, and user training to improve safe, trustworthy use.