How to use AI for work—without legal liability

In this TEDx talk, Constantine explores the real risks organizations face when adopting AI—risks grounded not in distant science-fiction scenarios but in the legal and operational realities unfolding today. Using the familiar metaphor of the unpredictable genie, he illustrates how generative AI interprets instructions literally, lacks human judgment, and can produce outcomes that expose organizations to significant liability.

Drawing from decades of privacy and governance experience, Constantine walks through real cases that demonstrate how quickly things can go wrong: a chatbot promising unauthorized discounts, confidential code uploaded into external training pipelines, and lawyers sanctioned for unverified AI-generated citations. Each example underscores a central point—the organization remains accountable for the actions and outputs of the tools it deploys.

The talk breaks down the three controls every organization must have to use AI responsibly: understanding the data that trains these systems, establishing effective governance and oversight, and ensuring staff are properly educated on both the capabilities and limitations of the technology. Constantine emphasizes that while organizations can outsource tasks, they cannot outsource liability. Without strong guardrails, AI becomes a source of hidden risk rather than meaningful efficiency.



Clear, practical, and grounded in real-world consequences, this talk challenges leaders to stop treating AI as a magical solution and start managing it with the discipline it requires.

Key Takeaways

• AI interprets instructions literally and can produce unintended outcomes.
  
  
• Organizations are legally responsible for the outputs of the AI tools they use.
  
  
• Poor-quality, biased, or unauthorized data creates immediate risk.
  
  
• Governance, policies, and vendor oversight are essential controls.
  
  
• Unapproved AI use inside organizations is common and often invisible.
  
  
• Staff must understand limitations, verification, and proper use.
  
  
• Effective AI adoption requires active oversight, not passive reliance.