Building AI Systems When Failure Is Not an Option

At TechCrunch Disrupt 2026, Shield AI's Nathan Michael, GM's Mikell Taylor, and Waabi's Raquel Urtasun discussed the challenges of building AI systems for the physical world, where failure can have severe consequences. A mistake could mean a grounded aircraft, a vehicle crash, or a compromised mission. The panel explored how to design AI systems that are robust, reliable, and safe. The conversation covered topics such as verification and validation, redundancy, and the importance of human oversight. The speakers shared their experiences from industries like defense, automotive, and autonomous trucking. They emphasized the need for rigorous testing and a safety-first mindset when deploying AI in high-stakes environments. This session is essential for anyone involved in building AI for critical applications. Follow TechCrunch for more from Disrupt 2026.