AI Safety & Responsible AI
Model leakage
Unintended disclosure of protected model behavior, parameters, or training information.
Example
Testing reveals memorized fragments of confidential training material in model output.
Why it matters
It helps teams investigate exposure arising from the model itself, not only connected files.
You'll hear it
“That reply appears to reproduce material that should have stayed private.”
Can you control it?
Not directly
No, not directly. Model leakage describes a learned property, observed result, or broader concept rather than a normal user setting. You may influence it indirectly by changing the tool, data, prompt, or workflow.
Common questions
- Does this require access to the original teaching files?
- No. A system may reveal learned information through its responses.
- Is model leakage identical to copying the entire model?
- No. Exposure can involve only fragments, behavior or other protected details.
- Can a copied passage prove every teaching source is accessible?
- No. One exposure does not establish what else can be recovered.