High-Stakes AI Deployment: How Defense, Autonomous Vehicles, and Robotics Handle Real-World Risk

The Challenge of Deploying AI Beyond the Lab

When artificial intelligence systems operate only in digital environments, mistakes can be corrected or dismissed. But when autonomous systems control aircraft, vehicles, or industrial robots, failures carry tangible consequences—crashes, mission compromises, safety breaches. This fundamental difference shapes how leading AI organizations approach development, testing, and deployment. The question driving industry conversation: how do you confidently move an autonomous system from controlled testing into real-world operations where human safety and critical missions depend on its performance?

Three prominent figures in high-stakes AI are bringing their expertise to this challenge. Nathan Michael, chief technology officer at Shield AI, leads development of Hivemind, a mission autonomy platform selected by the U.S. Air Force for its Collaborative Combat Aircraft program. Raquel Urtasun, founder and CEO of Waabi, spent 25 years in AI and autonomous vehicles, previously serving as chief scientist at Uber's autonomous vehicle division. Mikell Taylor leads robotics strategy for General Motors after spending over two decades building robots for practical, industrial-scale work, including leadership of Amazon's robotics team that developed the Proteus autonomous mobile robot.

The Defense Perspective: Autonomy Under Pressure

Shield AI represents the cutting edge of autonomous systems in defense applications. The company's Hivemind platform was selected as an autonomy provider for the Air Force's drone prototype program in February, validating the technology's readiness for military applications. This validation comes with significant market confidence—in March, Shield AI announced $1.5 billion in Series G funding, valuing the company at $12.7 billion post-money. Michael's background spanning AI, control, perception, and multi-robot systems—including his directorship of Carnegie Mellon University's Robotics Institute Resilient Intelligent Systems Lab—demonstrates the depth of expertise required to build systems that perform in unforgiving environments.

Defense applications demand not just high performance but assurance. A drone system must be reliable in contested environments, responsive to mission changes, and resilient to unpredictable conditions. The selection by the Air Force signals that Shield AI has achieved a level of maturity and trustworthiness that meets military standards.

Testing and Validation in Autonomous Vehicles

Waabi's approach to autonomous vehicles centers on rigorous validation before deployment. The company raised $1 billion in January and announced a partnership with Uber to deploy 25,000 or more robotaxis powered by Waabi's Driver technology. But before those vehicles operate without human drivers, Urtasun's team conducts extensive testing through Waabi World, a simulator that trains, tests, and stress-tests autonomous driving systems in virtual environments. This methodical approach reflects a fundamental principle: autonomous trucks require full validation before driverless deployment.

The stakes in autonomous driving are particularly visible because the general public will directly experience the technology. Unlike drone systems or warehouse robots operating in controlled industrial settings, robotaxis share roads with human drivers, pedestrians, and cyclists. The margin for error is measured not in mission success but in lives protected.

Building Robots People Trust and Work With

Taylor's robotics experience brings a human-centered perspective to autonomous systems deployment. Her work at Amazon on Proteus and her current role at General Motors reveal a critical insight: practical autonomous systems must be designed with user experience in mind from the beginning. Robots don't operate in isolation—they work alongside humans in factories, warehouses, and increasingly in shared spaces. Adoption depends not just on technical capability but on whether workers and operators can trust and effectively collaborate with these machines.

This principle differs subtly but importantly from the military and autonomous vehicle contexts. While all three domains require reliability, industrial robotics adds the dimension of human collaboration and workplace integration. A robot that works brilliantly in demonstrations but creates friction with human colleagues will fail in practice, regardless of its technical performance.

Common Ground Across Domains

Aircraft autonomy, autonomous vehicles, and industrial robots address fundamentally different challenges, but they converge on shared principles. All three require deep testing and validation frameworks. All three demand that safety cultures be embedded in organizations from engineering teams through leadership. All three face regulatory hurdles that vary by jurisdiction but consistently require demonstration of safety and reliability. And all three succeed only when they earn trust—from regulators, from operators, from the public.

The session brings these perspectives into direct conversation, allowing technology leaders and founders building autonomous systems to understand how established organizations in defense, transportation, and manufacturing approach the decisions between promising prototypes and systems ready for deployment. This comparison creates value beyond individual domains, as solutions from one field often illuminate challenges in others.