// MODEL OPTIMIZATION AND PROMPT SYNTAX TERM

Model Inversion

An attacker tries to figure out sensitive information about the data used to train an AI model by looking at its outputs.

TECHNICAL DEFINITION

A privacy attack where an adversary, given access to a trained AI model and potentially some output information (e.g., class probabilities), attempts to reconstruct or infer sensitive attributes of the training data instances, particularly those associated with a specific output.

BACKGROUND

Prompt engineering is the process of structuring natural language inputs to produce specified outputs from a generative artificial intelligence (GenAI) model. Context engineering is the related area of software engineering that focuses on the management of non-prompt contexts supplied to the GenAI model, such as metadata, API tools, and tokens.

READ MORE ON WIKIPEDIA

SYNONYMS & ALIASES

  • Data reconstruction
  • privacy leakage
  • attribute inference

USAGE NOTE

Model inversion attacks highlight the risk of exposing sensitive training data through model outputs.

DEVELOPERS

Organizations developing technology related to Model Inversion.

  • Google AI

    Engages in extensive research on privacy-preserving machine learning, including understanding and developing defenses against model inversion attacks to protect sensitive training data.

  • Microsoft Research

    Conducts deep research into AI security and privacy, publishing numerous papers and developing techniques to identify and mitigate various adversarial attacks on machine learning models, including model inversion.

  • IBM Research

    Focuses on trustworthy AI, developing tools like the Adversarial Robustness Toolbox (ART) which includes implementations of model inversion attacks and defensive strategies to enhance AI model security and privacy.

  • Meta AI (Facebook AI Research)

    Conducts fundamental and applied AI research, including significant work on the security and privacy aspects of machine learning models, exploring vulnerabilities like model inversion and developing countermeasures.

  • OpenAI

    Dedicated to ensuring advanced AI benefits humanity, their research includes AI safety, alignment, and understanding potential privacy vulnerabilities like model inversion in large language models to mitigate risks.

  • Private AI

    Develops software solutions for privacy-preserving machine learning, utilizing techniques like differential privacy and secure multi-party computation to protect sensitive data and prevent model inversion attacks.

  • Sarus

    Offers a privacy-preserving AI platform that enables secure data collaboration without sharing raw data, incorporating techniques to prevent data reconstruction and model inversion attacks.

  • Fortanix

    Provides confidential computing solutions that secure AI models and data in use, leveraging trusted execution environments to protect against various threats, including model inversion during inference and training.

RELATED TERMS IN AI ETHICS & SAFETY