Asimov's Three Laws of Robotics

Asimov's three rules of robot behavior, and their modern descendant, the rule sets large models use to keep robots safe.

Science fiction writer Isaac Asimov fully stated these in his 1942 short story “Runaround”: first, a robot may not injure a human being or, through inaction, allow a human being to come to harm; second, a robot must obey orders given by human beings, except where such orders conflict with the first law; third, a robot must protect its own existence as long as this does not conflict with the first two laws. A higher-priority “zeroth law” was added later: a robot may not harm humanity as a whole. The three laws were purely a fictional device. Now that large models are being connected to robots, Google DeepMind has borrowed the idea to propose a “robot constitution”: a set of safety rules written in natural language and placed in the prompt, so the model constrains itself when choosing tasks and deciding on actions. This supplements, rather than replaces, physical safety measures such as force limiting and emergency stops.

ExampleGoogle DeepMind's AutoRT (2024) writes a robot constitution inspired by the three laws into its prompt, ruling out tasks involving people, animals, sharp objects, or electrical appliances; the 2025 ASIMOV benchmark uses an automatically generated and iteratively revised constitution to evaluate semantic safety, reaching an alignment rate of up to 84.3%.

Also called

Robot Constitution

Related

Embodied Safety · Safety Filter · Functional Safety · Emergency Stop · AutoRT · Gemini Robotics

Sources

Three Laws of Robotics – Wikipedia
Shaping the future of advanced robotics (Google DeepMind, AutoRT)
Generating Robot Constitutions & Benchmarks for Semantic Safety (arXiv 2503.08663)

As of

2025-03