SESKStuff Everyone Should Know
SESK: Stuff Everyone Should Know · sesk.info/ai-overfitting-privacy-leakage

Why can overfitting increase training-data privacy risk?

Overfit models behave more differently on members and nonmembers.

Start playing free

  1. They can leak more membership information ✔
  2. They encrypt every example
  3. They stop returning probabilities

Why: Overfitting makes a model unusually tailored to its training examples. That behavioral gap can give a membership attacker a stronger signal about which records were used.

Huh, didn't know that: Overfitting is important, but model structure can also affect how much information leaks.

The dad joke
The overfit model remembered the homework and forgot the lesson.
Like this card?
Play the game

References

Answer: They can leak more membership information. Overfitting makes a model unusually tailored to its training examples. That behavioral gap can give a membership attacker a stronger signal about which records were used.

  1. Membership Inference Attacks against Machine Learning Models arxiv.org/pdf/1610.05820
    The more overfitted a model, the more it leaks
    Checked 2026-10-11.

Think this answer is wrong? Tap "Challenge this answer" on the card in the game and tell us why.

Test yourself in the game

Artificial IntelligenceA model scores great on its training data but badly on new data. What is that called?Artificial IntelligenceTraining error is far lower than test error. Which problem does that point to?Artificial IntelligenceWhat does dropout randomly remove during a training step?

That's stuff everyone should know.