Dictionary entry
model autonomy
also AI model autonomy or autonomy capability
Definition of model autonomy
-
: the capability of an AI model or model-based agent to carry out extended sequences of real-world actions, acquire resources, and adapt to obstacles with limited human instruction or intervention
- The system card rated model autonomy by testing whether the agent could obtain cloud compute and launch an inference service.
- A model can score highly on a coding benchmark without showing enough model autonomy to sustain an operation outside the test environment.
Origin & history
Model autonomy became a named frontier-risk category in OpenAI's 2023 Preparedness Framework and appeared in subsequent system cards. Early versions grouped long-horizon action, resource acquisition, self-exfiltration, autonomous replication, and AI research capabilities under the label. Later frameworks separated some of these into more specific categories while continuing to evaluate autonomy-related capabilities.
Editorial references
Sources & further reading
- Preparedness Framework (Beta)↗ Official Safety Framework · OpenAI · 2023
- OpenAI o1 System Card↗ Official System Card · OpenAI · 2024-12-05
- Preparedness Framework, Version 2↗ Official Safety Framework · OpenAI · 2025
Test yourself
Which of these is the meaning of model autonomy?
Cite this entry
"Model autonomy." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/model-autonomy/. Accessed 14 Oct. 2026.