// MODEL OPTIMIZATION AND PROMPT SYNTAX TERM

Model Security

Protecting AI models from attacks, misuse, and unauthorized access.

TECHNICAL DEFINITION

The discipline focused on safeguarding AI models against various threats, including adversarial attacks (e.g., evasion, poisoning), intellectual property theft (e.g., model extraction), and ensuring their integrity, confidentiality, and availability throughout their lifecycle.

BACKGROUND

Prompt injection is a cybersecurity exploit and an attack vector in which innocuous-looking inputs are designed to cause unintended behavior in machine learning models, particularly large language models (LLMs). The attack takes advantage of the model's inability to distinguish between developer-defined prompts and user inputs to bypass safeguards and influence model behaviour. While LLMs are designed to follow trusted instructions, they can be manipulated into carrying out unintended responses through carefully crafted inputs.

READ MORE ON WIKIPEDIA

SYNONYMS & ALIASES

  • AI security
  • robust AI
  • trustworthy AI
  • AI safety

USAGE NOTE

Model security is paramount for deploying reliable and safe AI systems in critical applications.

DEVELOPERS

Organizations developing technology related to Model Security.

  • Microsoft

    Develops comprehensive tools and practices for Responsible AI, MLOps security, and adversarial robustness, focusing on securing AI models throughout their lifecycle against various threats, including those from prompt manipulation.

  • Google

    Invests in secure and responsible AI development, offering platforms and research on model robustness, defenses against adversarial attacks, and interpretability to ensure the security of AI systems against malicious inputs.

  • IBM Research

    Conducts extensive research and develops solutions for AI trustworthiness, including robustness against adversarial attacks, fairness, and explainability, all critical for securing AI models and ensuring their integrity.

  • Adversa AI

    Specializes in AI security, offering solutions to detect and defend against adversarial attacks on AI models, thereby protecting model integrity and preventing manipulation through malicious inputs.

  • Robust Intelligence

    Provides an AI firewall and MLOps security platform designed to prevent AI failures, secure models from adversarial attacks, and ensure robust and reliable model behavior in production environments.

  • OpenAI

    Heavily invests in AI safety and security research, focusing on preventing misuse, ensuring model alignment, and mitigating vulnerabilities related to prompt engineering, such as prompt injection and data exfiltration.

  • Anthropic

    Dedicated to AI safety and interpretability, developing techniques and models like 'Constitutional AI' to make large language models more robust, steerable, and secure against various forms of adversarial prompting and misuse.

  • Trail of Bits

    A cybersecurity research and consulting firm that audits and develops tools to identify and mitigate security vulnerabilities in AI/ML systems, including issues related to model integrity, data privacy, and adversarial robustness.

RELATED TERMS IN AI ETHICS & SAFETY