// MODEL OPTIMIZATION AND PROMPT SYNTAX TERM
Model Extraction
An attacker tries to steal a copy of a trained AI model by querying it and observing its outputs.
TECHNICAL DEFINITION
An adversarial attack where a malicious actor attempts to reconstruct or replicate a target AI model's functionality, parameters, or architecture by observing its input-output behavior through queries, often to bypass intellectual property or security measures.
BACKGROUND
Artificial intelligence (AI) is the capability of computational systems to perform tasks typically associated with human intelligence, such as learning, reasoning, problem-solving, perception, and decision-making. It is a field of research in engineering, mathematics and computer science that develops and studies methods and software that enable machines to perceive their environment and use learning and intelligence to take actions that maximize their chances of achieving defined goals.
READ MORE ON WIKIPEDIASYNONYMS & ALIASES
- Model stealing
- model replication
- intellectual property theft
USAGE NOTE
Model extraction attacks pose a threat to the intellectual property of AI developers.
DEVELOPERS
Organizations developing technology related to Model Extraction.
Develops an end-to-end platform for AI safety and security, providing tools to detect and prevent adversarial attacks, including model extraction, ensuring the robustness and integrity of AI systems.
Offers an AI security platform designed to identify and mitigate vulnerabilities in machine learning models, protecting them from adversarial attacks such as model extraction, poisoning, and evasion.
Conducts leading research in AI security and privacy, including developing methods to understand and defend against model extraction attacks, contributing to the trustworthiness and resilience of AI systems.
Engages in fundamental and applied research in artificial intelligence, with significant efforts dedicated to AI security, privacy, and robustness, including techniques to protect machine learning models from intellectual property theft like model extraction.
A broad initiative spanning research and product development at Google, focusing on advancing AI capabilities while also addressing critical aspects like security, privacy, and responsible deployment, which includes strategies to prevent model extraction and intellectual property theft.
Conducts cutting-edge AI research with a strong emphasis on safety and security, developing techniques to protect their advanced language models from malicious use and various adversarial attacks, including the unauthorized extraction of model parameters or functionality.
An AI safety and research company dedicated to building reliable and steerable AI systems, with significant efforts in understanding and mitigating security risks, including the protection of proprietary model intellectual property from extraction attacks.
Develops comprehensive AI platforms and conducts research that contributes to securing AI workloads and models, including advancements in confidential computing and federated learning, which offer defenses against unauthorized access and potential model extraction.