
Mitigation Recommendations

Why securing AI is harder than anyone expected and the coming security crisis | Sander Schulhoff
Topic 12 of 14
Published Dec 21, 2025 · Clip 44:45–55:50 (11:06)
Sander advises: For simple chatbots, no defenses needed as damage is reputational and unavoidable. Ensure proper permissioning; users can force any action/data access. Hire AI security experts at the cyber-AI intersection. Use CAMEL to restrict agent permissions based on tasks. Education via courses like Maven's on red teaming and governance prevents poor deployments.