CyberArk Launches FuzzyAI: Revolutionizing AI Security
CyberArk, the global leader in identity security, has unveiled FuzzyAI, a groundbreaking open-source framework designed to identify and address vulnerabilities in AI models. This innovative tool is capable of jailbreaking every major tested AI model, providing organizations with critical insights into potential security weaknesses, including guardrail bypassing and harmful output generation. FuzzyAI is designed to support both cloud-hosted and in-house AI models, helping companies proactively manage cyber risks associated with AI deployment.
Why FuzzyAI?
As AI models continue to revolutionize industries, transforming customer interactions, improving internal processes, and enabling automation, they also introduce new security challenges. Unfortunately, most organizations are not fully prepared to address these emerging risks.
FuzzyAI offers a systematic solution by testing AI models against a range of adversarial inputs to uncover vulnerabilities in their security systems. By revealing flaws, FuzzyAI enables safer AI development and deployment. At its core, FuzzyAI features a powerful fuzzer, a tool designed to expose software defects and vulnerabilities. The fuzzer utilizes over ten distinct attack techniques, including bypassing ethical filters and exposing hidden system prompts, to ensure comprehensive security assessments.
Key Features of FuzzyAI:
- Comprehensive Fuzzing: FuzzyAI tests AI models using multiple attack techniques, uncovering vulnerabilities such as guardrail bypassing, information leakage, prompt injection, and harmful output generation.
- Extensible Framework: The tool is customizable, allowing organizations and researchers to add their own attack methods to test for domain-specific vulnerabilities.
- Community Collaboration: FuzzyAI benefits from a community-driven ecosystem, ensuring the continuous development of new adversarial techniques and defense mechanisms.
Commitment to AI Security
“The launch of FuzzyAI marks a significant step forward in securing AI models and addressing the inherent risks in AI model usage,” said Peretz Regev, Chief Product Officer at CyberArk. “Developed by CyberArk Labs, FuzzyAI has already demonstrated its ability to jailbreak every major tested AI model. This empowers organizations and researchers to actively identify weaknesses and strengthen their AI systems against evolving threats.”
Explore FuzzyAI at Black Hat Europe 2024
To learn more about how FuzzyAI can help mitigate AI-related cyber risks, attendees at Black Hat Europe 2024 will have the opportunity to explore its capabilities and applications firsthand.
