For Heads-of · Practitioner
Uplift
The additional ability an AI system gives a person to do something, compared with what they could do without it, for example using search engines and textbooks. Uplift is the measure most frontier safety thresholds use for dangerous capabilities.
- uplift studies
- dangerous capabilities
- CBRN
- cyber
- misuse
How it is measured
Uplift studies compare people doing a task with and without AI help, such as planning a laboratory procedure, writing exploit code or researching a target. If the group with AI help does meaningfully better, the model provides uplift. The comparison is against a realistic baseline, such as internet access, not against nothing.
Why it matters
A model that repeats information freely available elsewhere adds little risk; a model that helps a novice do what only an expert could is a different matter. Most frontier safety frameworks set their highest thresholds on uplift for chemical, biological, radiological and nuclear weapons and for cyber attack, and the Frontier Model Forum has published work on how such thresholds should be set.
For a deploying organisation
The idea applies well beyond weapons. Any AI system that makes it easier for someone to do harm, from writing convincing phishing messages to getting round a fraud control, provides uplift to misuse. Who could misuse a system, and how much easier it makes their task, belongs in the impact assessment; the register entry is adversarial testing coverage gap.
Sources
Related Terms
AI Evaluation
Systematic measurement of what an AI model or system can do, how it tends to behave and how well its safeguards hold, using benchmarks, task-based tests, red teaming or studies with people. Evaluations are the evidence most AI claims rest on, and they have known weaknesses.
AI Red Teaming
Structured adversarial testing in which testers try to make an AI system fail, either by producing harmful or prohibited output (safety red teaming) or by being subverted to act against its owner (security red teaming). It finds failures an ordinary test suite cannot, because the tester's job is to imagine what the designers did not.
Frontier AI
The most capable general-purpose AI models, at or near the leading edge. There is no single definition: the EU AI Act presumes systemic risk above 10^25 training operations, California's SB 53 and New York's RAISE Act define frontier models above 10^26, and the International AI Safety Report prefers the term general-purpose AI.
Frontier Safety Framework
A frontier AI developer's published commitment to test its models for dangerous capabilities and to apply stronger safeguards, or pause, when defined capability thresholds are crossed. Examples include Anthropic's Responsible Scaling Policy, OpenAI's Preparedness Framework and Google DeepMind's Frontier Safety Framework.