Security & Privacy
Model extraction
An attack that tries to copy a model's behavior through repeated queries.
Example
Repeated API queries help an attacker build a substitute model with similar behavior.
Why it matters
It helps identify attempts to reproduce a protected service without obtaining its original files.
You'll hear it
“These requests seem designed to map how the system behaves.”
Can you control it?
Set by the team
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Does extraction always recover the original internal values?
- No. A copied approximation may imitate behavior without reproducing the original design.
- Can a copy work well only on familiar cases?
- Yes. Matching observed responses may not reproduce broader abilities.
- Does a similar answer prove extraction occurred?
- No. Shared methods or common information can produce similar behavior independently.