As large language models (LLMs) become embedded across enterprise applications, the sophistication of AI-targeted attacks is growing just as fast. In response, Radware has introduced its LLM Firewall, a new security offering aimed at helping organisations safely accelerate their use of generative AI technologies.
LLM modules, once considered purely innovative, have quickly become a new attack surface. Prompt injection, jailbreaks, data exfiltration, and misuse of model resources are now among the most pressing concerns for CIOs and CISOs exploring enterprise AI adoption. Radware’s LLM Firewall is designed to counter these challenges by applying real-time, prompt-level protection that sits in front of the LLM itself.
“Businesses want to use AI, but they’re being held back by uncertainty around regulation, data protection, and system security,” said Constance Stack, Chief Growth Officer at Radware. “The LLM Firewall addresses these concerns head-on by stopping malicious prompts and model manipulation before they can take hold.”
Rather than relying on model-specific guardrails, Radware’s solution functions independently of the underlying LLM architecture. This model-agnostic design means organisations can protect AI interactions regardless of whether they run proprietary models, open-source engines, or third-party AI platforms.
Beyond blocking malicious behaviour, the LLM Firewall helps organisations prevent the unintended leakage of personal or sensitive data during LLM interactions an essential requirement for compliance with frameworks such as GDPR, HIPAA, and global data protection standards.
Radware confirms the new product is also designed to align with the 2025 OWASP Top 10 for LLMs and GenAI applications, helping enterprises mitigate the most prevalent AI-driven risks identified in the global security community.
The LLM Firewall is available as part of Radware’s Cloud Application Protection Services, complementing its wider suite of application, API, and infrastructure security tools that use AI-driven algorithms to provide real-time defence against sophisticated threats.
