AI Safety & Responsible AI
AI safety case
An evidence-backed explanation of why specified AI risks are adequately managed in a particular setting.
Example
A team gathers test results and assumptions to support a decision to use an AI.
Why people use it
It organizes the reasons and evidence behind a claim that an AI use is acceptably controlled.
What you'll hear
“What evidence supports the safety claim?”
What this means for you
Update the argument when the system or its use changes.
Can you control it?
Yes
Yes. A responsible team can prepare the argument, gather evidence and assign people to investigate unresolved questions.
Common questions
- Does a safety case prove the system has no risk?
- No. Its conclusions depend on scope, evidence and assumptions.
- Can evidence from one use cover a different use automatically?
- No. Different people, conditions or actions can change the relevant risks.
- Can the case become outdated?
- Yes. Changes to the system or its surroundings can weaken earlier evidence and assumptions.