<p>As AI becomes more deeply embedded in government and national security systems, one question grows harder to ignore: can these tools actually be secured? This week, Elisa is joined by Disesdi Shoshana Cox, CEO and Head of Research at Bermuda Hundred Strategies, to examine the cybersecurity risks behind AI integration. Together, they discuss whether AI models can be corrected after harmful inputs, how threat modeling can help secure national security use cases, and why prompt injections pose a serious risk as AI systems become more widely deployed across government.</p><p><a target="_blank" rel="noopener noreferrer nofollow" href="https://shoshana.cx/">Shoshana Cox</a> is the CEO and Head of Research at Bermuda Hundred Strategies and an AI security architect, researcher, and strategist whose work focuses on AI threat modeling, MLSecOps, and defensive architectures for mission-critical systems.</p><p><em>References:</em></p><ul><li><a target="_blank" rel="noopener noreferrer nofollow" href="https://doi.org/10.5281/zenodo.13905972">Cox, S. Securing AIML Systems in the Age of Information Warfare. Critical Alliance, Apr. 2022</a></li><li><a target="_blank" rel="noopener noreferrer nofollow" href="https://cloudsecurityalliance.org/artifacts/agentic-ai-red-teaming-guide#">Cox, S. <em>Agentic AI Red Teaming Guide</em>. Cloud Security Alliance, 28 May 2025</a></li><li><a target="_blank" rel="noopener noreferrer nofollow" href="https://disesdi.substack.com/">Shoshana's Substack</a></li><li><a target="_blank" rel="noopener noreferrer nofollow" href="https://disesdi.substack.com/subscribe?utm_source=menu&amp;simple=true&amp;next=https%3A%2F%2Fdisesdi.substack.com%2F">Subscribe to Angles of Attack Newsletter</a></li></ul>