WitnessAI on Wednesday rolled out two new products designed to address growing risks associated with enterprise AI deployments.
Attack and Protect: red-teaming and firewall defenses for AI security
The new tools, Witness Attack and Witness Protect, will function as an automated red-teaming tool and a runtime AI firewall, respectively.
The company’s CEO, Rick Caccia, framed the releases as part of a wider effort to reduce fragmentation in AI security.
“Enterprises don’t want to buy five different products to ensure their employees and customers can use AI safely,” he said in a press release.
Pressure-testing LLMs
LLMs are increasingly targeted by prompt injections, jailbreaks, and other manipulations that can expose sensitive data or generate harmful content. A 2024 study by some Cornell University researchers found that 56% of prompt injection attacks succeeded against major LLMs.
As a result, industry researchers and regulators have raised concerns that without proactive defenses, enterprises risk deploying models that cannot withstand even basic adversarial testing.
Witness Attack promises to address this issue through automated red-teaming. The company said this solution can simulate multimodal and multi-step attack methods, including fuzzing and reinforcement learning, to pressure-test model defenses before deployment. It claims that such tests give enterprises early visibility into potential flaws that could otherwise surface during live use.
A runtime firewall for live model protection
The company describes Witness Protect as an AI firewall that provides behavioral defenses once models are live.
The system applies filters, response controls, and real-time redaction to block malicious prompts and unauthorized requests. WitnessAI claims to detect more than 99% of prompt injection attempts, citing two years of training on synthetic attack data.
The company stated that the firewall supports over 100 different LLMs and provides enterprises with a standardized method of protection across various platforms. It added that the product has already been deployed in evaluations to replace existing AI firewall solutions.
Enterprise AI portfolio broadens at witnessAI
Founded in 2023 and backed by Ballistic Ventures and GV, WitnessAI has spent the past year broadening its enterprise AI security portfolio.
Last April, the Mountain View-based firm released version 2.0 of its Secure AI Enablement Platform, introducing controls that map to the Payment Card Industry Data Security Standard (PCI DSS) compliance. The company recently announced a partnership with TENEX.AI to provide managed security and compliance services for Enterprise AI.
The latest addition of automated testing and runtime firewalls suggests the company is seeking to establish a single framework for enterprises navigating AI development, compliance, and security challenges.