NSS Labs launches AI protection test

AI security tools sit outside a model like a checkpoint at a building door, screening prompts, outputs, and tool calls before damage spreads. NSS Labs said April 28 it has released a test method for those systems. (nsslabs.com 1) (nsslabs.com 2) The company calls the category AI Protection Systems, or AIPS: products that enforce policy, monitor interactions, and try to block misuse around a model, app, or agent. The new methodology is meant to measure those controls in enterprise deployments, not just score a raw model in isolation. (nsslabs.com 1) (nsslabs.com 2) NSS Labs said the method covers eight dimensions: prompt injection resistance, data exfiltration prevention, system resilience, policy enforcement accuracy, agentic AI and tool invocation security, observability and auditability, performance and scalability, and cross-model integration. (nsslabs.com) The attack set is built to look like real misuse, not a classroom quiz. NSS Labs said it runs hundreds of thousands of variations, including obfuscated prompts, role-play manipulation, instruction hijacking, retrieval-augmented generation poisoning, exploit-generation requests, and application programming interface misuse. (nsslabs.com) That reflects the problem enterprises are trying to solve. In a March 18 paper release, NSS Labs said model-level safety alone is not enough when companies connect AI to internal data, workflows, and delegated actions. (nsslabs.com) The company has been building toward this launch for weeks. On March 23, NSS Labs said Keysight Technologies became the lead partner in its AIPS security testing initiative, which it described then as one of the first independent programs focused on AI guardrail technologies. (prnewswire.com) NSS Labs also says the framework is not a single checklist every vendor must pass. Tests are aligned to each product’s documented features, and the results are meant to show which protections a product actually supports and where gaps remain. (nsslabs.com) That vendor-by-vendor framing fits NSS Labs’ broader return to the market. The company relaunched on July 9, 2025 under founder and chief executive Vikram Phatak after the original NSS Labs ceased operations in 2020. (prnewswire.com) The pitch to buyers is familiar from the company’s older firewall and endpoint work: independent testing, repeatable methods, and evidence that a product works under pressure. This time, the pressure point is AI systems that can leak data, ignore policy, or misuse connected tools at machine speed. (prnewswire.com) (nsslabs.com) NSS Labs’ message is that AI guardrails should be tested like any other security control: against the attacks they will actually face. The company says this methodology is the basis for that kind of proof. (nsslabs.com)