Waymo's experience deploying robotaxis safely is reshaping how the entire AI industry approaches evaluation and risk management, as companies discover that the discipline required to put self-driving cars on real streets offers crucial lessons for managing AI safety across voice assistants, chatbots, and other high-stakes applications. As artificial intelligence systems become more powerful and autonomous, the robotaxi industry's hard-won expertise in continuous monitoring, real-world testing, and human oversight is becoming the template for responsible AI development.

What Can AI Companies Learn From Autonomous Vehicle Safety?

The autonomous vehicle industry has long operated under intense scrutiny because the stakes are literally life and death. Every robotaxi deployment requires extensive testing, real-world monitoring, and continuous human oversight. Now, as AI safety concerns mount across the industry, experts are arguing that this same disciplined approach should become standard practice for all AI development.

Brooke Hopkins, who previously led safety evaluation at Waymo and now runs an AI observability company called Coval AI, explained the core problem: companies are moving faster than they can identify failures and understand when their systems can no longer be trusted. She noted that the speed at which AI systems evolve makes it even more critical that evaluation keeps pace with development.

"Companies can't rely on one type of test to confirm whether a system is ready for the real world. They need simulation, controlled environments, edge-case testing and real-world monitoring, followed by a gradual rollout," stated Hopkins.

Brooke Hopkins, Founder of Coval AI and former Waymo safety lead

This multi-layered testing approach is exactly what Waymo has refined over years of robotaxi operations. The company completes over 500,000 paid autonomous trips across 15 major American cities per week, generating real-world data that informs continuous safety improvements. That operational scale provides a natural testing ground that no laboratory simulation can fully replicate.

Why Do Current AI Evaluation Methods Fall Short?

The problem with most AI evaluation today is that it treats safety as a final checkpoint rather than an ongoing process. Companies run models through benchmark tests at launch, declare them safe, and deploy them. But benchmarks only measure how a system performs on specific tests at specific moments in time. They don't reveal how that system will behave across millions of real interactions, in unexpected situations, or when users deliberately try to break it.

Hopkins emphasized that what's missing is a much stronger focus on real-world behavior. She outlined the key gaps in current evaluation practices:

• Continuous telemetry: Real-time monitoring of how systems actually perform in production, not just in controlled lab environments
• Adversarial testing: Deliberate attempts to find failure modes and edge cases that standard benchmarks miss
• Human-in-the-loop oversight: Keeping people actively reviewing outputs and stepping in when something looks wrong, rather than relying solely on automated test suites

The autonomous vehicle industry learned this lesson the hard way. Simulation and controlled testing are necessary but insufficient. Real-world deployment, combined with continuous monitoring and human judgment, is what actually reveals whether a system is safe.

How to Build Safety Into AI Development From the Start

• Integrate evaluation throughout development: Rather than treating safety as a gate at the end of the process, embed testing and evaluation into every stage of building the AI system
• Set clear standards before launch: Define what success and failure look like upfront, then test against those standards consistently
• Roll out systems progressively: Deploy to limited audiences first, monitor real-world behavior, and expand only when confidence is high
• Maintain continuous measurement: Keep measuring what happens in production after launch, and be ready to intervene, fix issues, or roll back if systems start failing in ways that matter

Hopkins stressed that speed and safety don't have to be in conflict. In fact, the faster AI systems evolve, the more important it becomes that evaluation keeps pace. The key is building evaluation into the development process rather than treating it as a separate final step.

This approach mirrors what Waymo does with its robotaxis. The company doesn't deploy a new version of its autonomous driving software to all vehicles simultaneously. Instead, it tests extensively in simulation, validates in controlled environments, monitors edge cases, and then rolls out gradually while continuously measuring real-world performance.

What Would Effective AI Regulation Actually Look Like?

As concerns about AI safety intensify, regulators are beginning to ask what guardrails should look like. Hopkins argued that the most useful regulation would focus on accountability and outcomes rather than prescribing exactly how companies should build their technology.

For higher-risk systems, she suggested that companies should be required to understand and document the risks, report significant incidents, and be transparent about how they evaluate their systems and where those systems have limitations. For the most capable or high-impact AI systems, there should be a higher bar for independent evaluation and ongoing monitoring.

The goal shouldn't be to slow down innovation, but rather to ensure that when a company puts a powerful system into the world, there's a clear understanding of what it can do, where it can fail, and who is responsible when something goes wrong. Industry-specific standards matter too, since the risks and appropriate safeguards look very different for a voice assistant than for an autonomous vehicle.

Waymo's experience demonstrates that rigorous safety practices don't prevent innovation; they enable it. The company has deployed thousands of robotaxis across multiple cities precisely because it invested heavily in evaluation, monitoring, and human oversight. As AI systems become more autonomous and more consequential, the rest of the industry is learning that Waymo's playbook isn't just good practice for self-driving cars. It's becoming the template for responsible AI development across every domain.