Zero-Knowledge Proof (ZKP)

A cryptographic scheme enabling computation directly on ciphertext. An untrusted party can perform operations on encrypted data, generating an encrypted result that, when decrypted, matches the output of operations performed on the plaintext. This allows AI models to be trained or run inference on sensitive data without ever decrypting it.

• Fully Homomorphic Encryption (FHE): Supports arbitrary computation but is computationally intensive.
• Partially Homomorphic Encryption (PHE): Supports only specific operations (e.g., addition) but is more practical for targeted use cases.
• Use Case: A cloud provider trains a diagnostic model on encrypted medical images without accessing the raw patient data.

A cryptographic protocol that distributes a computation across multiple parties where no single party can see the others' private inputs. Each participant holds a fragment of the data, and the protocol computes a function over the combined inputs without revealing them.

• Secret Sharing: The underlying technique splitting data into meaningless shares distributed to different parties.
• Garbled Circuits: A method for secure two-party computation of Boolean circuits.
• Use Case: Multiple banks collaboratively train a fraud detection model without exposing their proprietary transaction records to competitors.

A mathematical framework that injects calibrated statistical noise into query results or model training to guarantee that the presence or absence of any single individual's data is indistinguishable in the output. The privacy loss is quantified by the parameter epsilon (ε).

• Epsilon (ε): The privacy budget; lower values provide stronger privacy guarantees.
• Global vs. Local DP: Global DP adds noise to the aggregated output; Local DP adds noise on each individual's device before collection.
• Use Case: Apple and Google use Local Differential Privacy to collect usage statistics from devices without learning about any specific user.

A hardware-enforced isolated area within a main processor that protects code and data loaded inside it with respect to confidentiality and integrity. Unlike purely cryptographic methods, TEEs rely on hardware root of trust.

• Intel SGX/TDX: Intel's implementation providing enclaves for confidential computing.
• AMD SEV: AMD's Secure Encrypted Virtualization for protecting virtual machines.
• Use Case: A model owner deploys proprietary AI inference code to an untrusted edge device, with the TEE ensuring the model weights and input data remain inaccessible to the device operator.

A decentralized machine learning technique where a shared global model is trained across multiple edge devices or servers holding local data samples. Only model weight updates—not raw data—are transmitted to a central aggregation server.

• Federated Averaging (FedAvg): The standard algorithm where local stochastic gradient descent updates are averaged to update the global model.
• Secure Aggregation: A cryptographic protocol ensuring the central server can only see the aggregated sum of updates, not individual contributions.
• Use Case: A keyboard app improves next-word prediction by training on user typing patterns locally, sending only encrypted model updates back to the central server.

A hardware-based security paradigm that protects data in use by performing computation within a hardware-based Trusted Execution Environment. It shields sensitive data from the operating system, hypervisor, and cloud provider, providing attestation that the environment is genuine.

• Attestation: The process by which a TEE proves to a remote party that it is running specific, untampered code on genuine hardware.
• Confidential VMs: Full virtual machines running within a TEE, lifting existing applications into a confidential environment without code changes.
• Use Case: A consortium of hospitals runs a joint analysis on combined patient data in a confidential VM, with cryptographic proof that only the agreed-upon analysis code executes.