How Shield AI, Waabi, and GM Ensure AI Safety in High-Stakes Environments

Overview

As artificial intelligence transitions from digital assistants to physical actors, the margin for error disappears. When AI controls aircraft, autonomous vehicles, or industrial robots, a single hallucination or sensor glitch can result in catastrophic physical consequences rather than mere informational errors.

This fundamental shift is the focus of a critical panel at TechCrunch Disrupt 2026. The session, titled Building AI Systems When Failure Is Not an Option, brings together leaders from defense, autonomous driving, and industrial robotics to discuss validation, safety cultures, and regulatory compliance.

When artificial intelligence moves into the physical world, a mistake can ground an aircraft, cause a vehicle crash, or compromise a mission.

The discussion takes place on the Real World AI Stage, featuring Nathan Michael of Shield AI, Waabi founder and CEO Raquel Urtasun, and General Motors’ Mikell Taylor. Their collective expertise highlights the rigorous standards required to deploy autonomous systems in high-stakes environments.

Defense and Aerospace: Assurance Through Redundancy

Nathan Michael, Chief Technology Officer at Shield AI, brings extensive experience in multi-robot systems and control theory to the conversation. His background includes directing the Resilient Intelligent Systems Lab at Carnegie Mellon University’s Robotics Institute, where he focused on systems that must operate reliably in unpredictable conditions.

At Shield AI, Michael leads the development of Hivemind, a mission autonomy software platform designed to be agnostic across various hardware systems. This technology is not confined to laboratory settings; it has been selected as an autonomy provider for the U.S. Air Force’s Collaborative Combat Aircraft drone prototype program.

The stakes in defense are exceptionally high, requiring a balance between advanced performance and absolute assurance. Shield AI’s recent $1.5 billion Series G funding round, which valued the company at $12.7 billion, underscores the market’s confidence in its ability to deliver reliable autonomy.

Michael’s perspective at Disrupt will address how developers build trust in AI when the cost of failure is measured in national security and human lives. The emphasis is on creating systems that can withstand adversarial conditions and unexpected environmental variables without compromising mission integrity.

Autonomous Driving: Simulation as the Gateway to Safety

Raquel Urtasun’s approach to AI safety centers on the premise that real-world testing alone is insufficient for validating fully autonomous systems. With 25 years of experience in AI and autonomous vehicles, including her tenure as chief scientist at Uber ATG, Urtasun understands that thousands of miles of driving are necessary to prove safety.

Waabi, the company she founded, has developed Waabi World, a high-fidelity simulator that allows the company to train, test, and stress-test its autonomous driving stack in virtual environments. This simulation-first methodology enables rapid iteration and exposure to rare, dangerous edge cases that would be impractical to encounter in physical testing.

In January, Waabi announced a $1 billion funding round and a strategic partnership with Uber to deploy over 25,000 robotaxis powered by Waabi Driver. However, Urtasun maintains that the technology still requires comprehensive validation before full driverless deployment on public roads.

The central question for Waabi is not just whether the car can drive, but whether it can drive as safely as, or better than, a human operator. Urtasun argues that rigorous simulation and validation frameworks are the only viable path to achieving the statistical confidence needed for mass adoption.

How do you determine when an AI system is ready to operate without a human behind the wheel?

Industrial Robotics: Designing for Human Collaboration

While defense and automotive sectors focus on isolation and redundancy, General Motors approaches safety through the lens of human-robot interaction. Mikell Taylor, Director of Robotics Strategy for GM’s Autonomous Robotics Center, emphasizes that utility and reliability must be designed alongside user experience.

Taylor’s career began with unconventional projects, including building a robotic prom date, but quickly evolved to lead the Amazon Robotics team that developed Proteus, Amazon’s first autonomous mobile robot. Her work spans autonomous underwater vehicles and industrial systems designed to operate alongside human workers.

At General Motors, Taylor is responsible for ensuring that robotics solutions are practical, reliable, and safe for factory environments. The challenge here is distinct: the AI does not need to navigate complex traffic laws, but it must predict human behavior and respond appropriately in shared workspaces.

Taylor advocates for a design philosophy where safety and adoption are considered from the outset. This involves understanding how humans interact with autonomous machines and ensuring that these systems can be trusted in dynamic, populated environments.

The Convergence of Safety and Trust

The panel at TechCrunch Disrupt 2026 illustrates that while the domains differ, the underlying challenge is identical: how to certify readiness when failure is unacceptable. Whether flying a drone in contested airspace, driving a robotaxi through a busy city, or working alongside humans in a factory, the consequences of AI error are physical and immediate.

Key themes emerging from these leaders include:

1. Validation Rigor: Moving beyond basic functional testing to stress-test systems against rare and extreme scenarios.
2. Simulation Fidelity: Using virtual environments to accelerate training and expose systems to dangers before they encounter them in reality.
3. Human-Centric Design: Ensuring that autonomous systems are not only technically proficient but also predictable and trustworthy to the humans around them.
4. Regulatory Navigation: Proactively engaging with regulatory bodies to establish safety standards that do not stifle innovation but ensure public confidence.

For founders and technology leaders, the session offers a comparative look at how different industries tackle the spectrum between promising prototypes and deployed reality. It addresses the critical gap that exists when AI leaves the controlled environment of the lab and enters the unpredictable physical world.

What does it take to trust AI with a mission?

Attendees will have the opportunity to hear directly from Michael, Urtasun, and Taylor as they dissect the engineering, ethical, and operational decisions that define safe AI deployment. The insights gained from these discussions are vital for any organization looking to move beyond experimental AI into robust, real-world applications.