Safety
AI safety
Work to prevent AI systems from causing harm, through misuse, accidents or behaviour their developers did not intend.
Safety spans testing models for dangerous capabilities, training them to refuse harmful requests, monitoring deployments and research into making powerful systems controllable.
Frequently asked questions
What does AI safety mean?
Work to prevent AI systems from causing harm, through misuse, accidents or behaviour their developers did not intend. Safety spans testing models for dangerous capabilities, training them to refuse harmful requests, monitoring deployments and research into making powerful systems controllable.