Dangerous Capability Evaluation
A dangerous capability evaluation tests whether an AI model can perform tasks that could cause severe harm, such as helping create biological weapons, conducting cyber attacks or replicating itself autonomously.
What is Dangerous Capability Evaluation?
These evaluations are central to frontier developers' safety frameworks and to the work of government AI safety and security institutes. Results determine which safeguards are required before deployment.
How it works
Experts design tasks representing harmful capabilities and measure the model's success, sometimes compared with human experts using the internet alone.
← All Artificial Intelligence & Superintelligence definitions · ← Back to the A–Z definitions