Secure Inference

Intermediate

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

Full Definition

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

Keywords

TEEs enclaves

Domains

Foundations & Theory

Related Terms

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.

 Memory related to

Mechanisms for retaining context across turns/sessions: scratchpads, vector memories, structured stores.

 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.

 Privacy Attack related to

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

Concept Map

See how Secure Inference connects to other concepts.

Open Knowledge Graph