Automated red-teaming helps protect enterprise AI
Artificial intelligence is being used to streamline many business tasks, but at the same time it opens up new attack vectors and risks.
Secure AI specialist WitnessAI is announcing two new products aimed at securing enterprise LLMs and AI applications through automated red-teaming and behavioral runtime protection.
Witness Attack is an automated red-teaming tool designed to proactively identify vulnerabilities in large language models (LLMs) before they are deployed. It uses simulated attack techniques such as multimodal attacks, multi-step jailbreaks, comprehensive fuzzing, API integration, and reinforcement-learning attacks, to generate synthetic prompts that stress-test the model's defenses. With Witness Attack, developers can easily identify and address weak spots in their models’ own defenses before they are rolled out.
Witness Protect is a next-generation AI firewall that provides behavioral runtime defense against model attacks such as prompt injections and jailbreaks. It also provides LLM defenses such as behavioral prompt filtering, intention-based response control, real time data redaction, threat detection, harmful response, toxicity filtering, and model identity enforcement, all delivered through a simple developer API.
Witness Protect includes new defense techniques developed from two years’ worth of synthetic prompt data and continuous conversational attacks. As a result, the solution provides industry-leading detection of model attacks, with over 99 percent effective identification of prompt injections. Witness Protect’s defenses provide standardized protection across models, and work with over 100 types of LLMs.
“Enterprises don’t want to buy five different products to ensure their employees and customers can use AI safely,” says Rick Caccia, WitnessAI’s CEO. “With the introduction of Witness Attack, enterprises can now ensure automated testing and hardening of their internally-developed models, apps, and agents. Witness Protect adds even better defenses against model attacks, and is already in customer evaluations to replace previously-deployed AI firewall solutions from legacy security providers.”
You can find out more on the WitnessAI site.
Image credit: Mediaphotos/Dreamstime.com