Privacy Attack

Intermediate

Attacks that infer whether specific records were in training data, or reconstruct sensitive training examples.

Full Definition

Attacks that infer whether specific records were in training data, or reconstruct sensitive training examples.

Keywords

membership inference

Domains

Foundations & Theory

Related Terms

Secure Inference related to

Methods to protect model/data during inference (e.g., trusted execution environments) from operators/attackers.

 Human-in-the-Loop related to

System design where humans validate or guide model outputs, especially for high-stakes decisions.

 Automation Bias related to

Tendency to trust automated suggestions even when incorrect; mitigated by UI design, training, and checks.

 Planning related to

Methods for breaking goals into steps; can be classical (A*, STRIPS) or LLM-driven with tool calls.

 Prompt Injection related to

Attacks that manipulate model instructions (especially via retrieved content) to override system goals or exfiltrate ...

 Data Poisoning related to

Maliciously inserting or altering training data to implant backdoors or degrade performance.

 Backdoor / Trojan related to

Hidden behavior activated by specific triggers, causing targeted mispredictions or undesired outputs.

 Model Stealing related to

Reconstructing a model or its capabilities via API queries or leaked artifacts.

Concept Map

See how Privacy Attack connects to other concepts.

Open Knowledge Graph