Trust in Robotics: From Čapek’s Robots to AI Building AI

By Dorian Cartwright | TrustRobotics™

Mistrust accompanied the word “robot” from its beginning.

In Karel Čapek’s 1920 play R.U.R.—Rossum’s Universal Robots, humans manufacture artificial workers to serve them. The robots eventually revolt and nearly extinguish humanity. The story invites a question that remains relevant: what justified humanity’s confidence that its creations would remain under its control? Background on R.U.R.

AI building AI: the modern warning

More than a century later, a similar question is coming from the people building artificial intelligence.

In his September 2026 essay, We Must Pace the Frontier, Anthropic CEO Dario Amodei warns that AI’s growing role in building the next generation of AI could outpace human understanding and control. He calls for slower capability advancement, independent evaluators, and coordinated safety measures. His concern includes commercial competition pushing development faster than safeguards can mature. Dario Amodei’s essay

Anthropic’s companion analysis, When AI Builds Itself, explains the mechanism: AI helps develop better AI, which can then contribute to further development. It also makes an essential qualification: fully autonomous development of successor systems has not yet been achieved, and is not inevitable. Anthropic’s analysis

Creating intelligence does not establish control

Čapek dramatized a rebellion. Anthropic describes a possible failure of oversight during accelerating technological development. Those are different mechanisms, but they expose a shared assumption: that the power to create something includes the power to keep directing it.

When intelligence becomes physical action

For robotics, this question acquires a physical dimension. An AI system can propose a change in software. A humanoid can apply force, move equipment, enter a room, or interact with a person. If AI also helps design robot software and future robot capabilities, trust must extend across both development and deployment.

Who authorizes a new capability? What evidence supports allowing it to operate? Can its permissions be withdrawn? What prevents a learned behavior from producing an unacceptable physical action?

Making trust demonstrable

These are the questions that motivate TrustRobotics™. Our focus is on making physical execution subject to explicit authority, applicable constraints, and evaluation of present conditions. A robot’s ability to plan an action should be accompanied by a dependable process for determining whether that action may proceed.

Such controls are one part of the larger challenge. They do not, by themselves, solve the risks of AI developing AI. But they address a concrete point at which intelligence produces consequences: the moment a proposed action becomes physical execution.

Čapek’s story ended in humanity losing control. Our responsibility is to build systems in which continued human control is something we can demonstrate.