New offering enables customers to independently evaluate AI guardrails and gateways outside of lab benchmarks to better protect their systems with data-informed security decisions
Mindgard, the leader in AI security, released GuardBuster, a new offering which brings together Mindgard’s platform, research, and adversarial AI security expertise to evaluate the effectiveness of AI guardrails and gateways under realistic and agentic attack conditions. As enterprises deploy AI systems, agents, copilots, and LLM-powered applications, many are turning to guardrails as a first line of defense against prompt injection, jailbreaks, and data leakage. However, most guardrails are built and benchmarked in controlled lab environments, resulting in a lack of independent evidence around how those protections perform against adaptive, real-world threats.
Also Read: CIO Influence Interview with Kyle Wickert, Field CTO at AlgoSec
Due to narrow or vendor-influenced testing scenarios, enterprise AI buyers and builders do not have the expertise or tooling to determine whether guardrail benchmarks are actually effective, resulting in the need for frequent re-evaluation, as its context and environment are constantly changing. Vendor-reported accuracy rates can create confidence, but that confidence may not fully reflect real-world attack conditions. If a guardrail performs well against known benchmarks, but fails against adaptive attackers, an organization may have an unwarranted assurance of reduced risk while significant exposure remains.
Available now, Mindgard’s GuardBuster helps organizations evaluate defenses against adaptive adversarial behavior rather than static, familiar benchmark prompts. The tool is designed to test how guardrails perform when exposed to more realistic adversaries, leveraging a variety of techniques, including psycho-analytical coercion, subtle prompt injection and jailbreaking, character-level evasion, adversarial machine learning evasion, and multi-turn manipulation, contextual obfuscation, amongst other means.
“If an organization invests in a guardrail, but cannot measure it effectively, they’re facing a gap that still must be addressed,” said Aaron Portnoy, Chief Product Officer at Mindgard. “The AI ecosystem needs independent validation that shows not just whether a control passes or fails, but what type of attacks it can stop, how systems respond under adversarial pressure, and where defenses begin to break down. With this offering, Mindgard acts as the complement to any guardrail, enabling organizations to validate their security investments with proven value, and empowering customers to push back on vendors who aren’t delivering quality assessments.”
Mindgard research confirms that LLM guardrail systems exhibit major blind spots and limitations against real-world attacks, and unveils the evasive nature of prompt injection and jailbreak detection systems and how vulnerable current LLMs are to these threats. As AI security evolves beyond testing models in isolation, independent assessment is becoming a necessity to ensure enterprise AI applications can withstand today’s motivated attackers. Rather than relying solely on a vendor’s claim, there is an immediate need for organizations to conduct an independent analysis and test continuously as attacks evolve, not only to confirm security, but to harden products and actually reduce risk in their own applications.
Guardrails are important, but buyers and builders need credible evidence that they work under realistic adversarial conditions. With GuardBuster, Mindgard addresses the growing need for organizations to assess how AI defenses perform in production environments, where attacks are dynamic, adaptive, and capable of targeting far more than the model alone.
Catch more CIO Insights: The CIO as a Value Creator: Moving Beyond Cost Centers to Revenue Drivers
[To share your insights with us, please write to psen@itechseries.com ]
