Security & Privacy
Data poisoning
Corrupting training data so a model learns harmful, biased, or attacker-chosen behavior.
Example
Attackers insert misleading examples into a training collection to corrupt later predictions.
Why it matters
It helps teams recognize attacks aimed at changing what a system learns.
You'll hear it
“These new examples are trying to teach the wrong association.”
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 poisoning have to affect most of the data?
- No. Carefully placed examples can sometimes influence particular behavior.
- Can the system look normal on ordinary tests?
- Yes. A targeted effect may appear only in selected circumstances.
- Is accidental poor data the same as deliberate poisoning?
- No. The consequences may overlap, but poisoning implies intentional interference.