Nol8 launches AI Data Plane with 1 GB/s deterministic inspection

Nol8 launched its AI Data Plane for commercial use on November 13, 2026, a platform that inspects data flowing into and out of AI models, agents and tools and claims to sustain 1 GB/s throughput while policy sets grow from 128 rules to 5,000.

The launch targets a problem that has moved from the margins of enterprise IT to the center: as companies shift from testing AI tools to deploying large fleets of autonomous agents, they need to control what sensitive information reaches those systems during routine tasks. Nol8's answer is a data plane that sits between enterprise data and AI workloads, reviewing prompts, retrieved documents, tool calls, agent-to-agent exchanges and model outputs, then applying company-defined rules to decide what passes through.

The company, a business unit of FortifAI, says the platform is available immediately on AWS for production deployments.

What did the benchmarks find?

Benchmark results released alongside the launch compared Nol8's system against a CPU-based baseline built on Google RE2, an open-source pattern-matching engine. Nol8 says an independent party validated the testing.

Across the policy range examined, the company reported:

• More than 100 times more data processed per dollar than the RE2 baseline at the top end of the policy range
• Sustained throughput of 1 GB/s as policy size grew from 128 rules to 5,000 rules
• A 30-fold slowdown for the RE2 baseline as rule volumes increased

Nol8 argues these numbers matter because the cost of inspecting AI traffic rises sharply as organizations add policies. Inspection economics are becoming part of the AI adoption debate: a governance layer that slows down or grows expensive with each new rule pushes companies toward either weaker controls or sampled inspection.

Why is deterministic matching the design choice?

Nol8's technical approach departs from the probabilistic controls common in AI safety tooling. The platform uses deterministic pattern matching: the same rule produces the same result on every request. If the platform becomes unreachable, requests are blocked rather than allowed through — a fail-closed posture aimed at compliance guarantees.

Dr. Alon Rashelbach, Co-Founder and Chief Technology Officer of Nol8, framed determinism as a structural requirement rather than a preference.

"Data governance for AI must be deterministic. If a policy says a class of data cannot reach a model, that must hold on every request, not most of them," Rashelbach said.

"The NOL8 AI Data Plane matches patterns rather than inferring intent, so its latency barely moved as the policy grew. Security and compliance teams can control the data AI systems see without engineering teams paying for it in latency or infrastructure sprawl," he added.

The distinction carries weight for regulated industries. Probabilistic classifiers can miss novel formats of sensitive data or behave inconsistently across requests. Pattern matching gives an auditable rule set, at the cost of requiring policy authors to define what counts as sensitive rather than relying on a model to infer intent.

What problem does the CEO say enterprises face?

Kelly Herrell, Chief Executive Officer of Nol8, said the governance conversation has focused on the wrong layer of the stack. In his framing, data access precedes model behavior as the first question an enterprise must answer.

"Most of the conversation about enterprise AI has focused on whether it behaves correctly. Before behavior comes a more basic question: what data does AI actually access? AI needs broad access to be useful, and that access expands every day as AI pushes deeper into the business," Herrell said.

"With NOL8, companies predetermine what data can and cannot pass into and out of AI systems and enforce it at machine speed, without slowing autonomous workloads. They can give AI the access it needs and still decide exactly what data it receives. That certainty is now available to any enterprise running AI on its data," he said.

The position reflects a practical tension in enterprise deployments. Agents that retrieve documents, call tools and exchange information with other agents need broad data access to be useful. The same breadth of access is what exposes sensitive information to models during ordinary operations. Nol8's pitch is that enterprises should not have to choose between useful AI and controlled AI.

How big is the agent wave the platform is betting on?

Industry forecasts cited by Nol8 point to steep growth in autonomous agent deployments over the next two years.

• Gartner has forecast that by 2028 the average global Fortune 500 company will run more than 150,000 AI agents, up from fewer than 15 in 2025.
• IDC has projected a tenfold increase in agent use across the Global 2000 by 2027.

If those projections hold, the volume of prompts, retrievals, tool calls and agent handoffs that enterprises must inspect will grow by orders of magnitude — and per-request inspection costs, latency and policy complexity will scale with it. That scale argument is the core of Nol8's commercial case.

The platform targets enterprises, inference providers and colocation operators that want predictable costs for AI data inspection as policy requirements expand.

Why target data in motion rather than data at rest?

Nol8's design breaks from traditional data governance systems, which center on stored information: databases, file shares and data catalogs protected by static policies. The new platform instead targets what the company describes as data in motion — the live flow of information through prompts, context retrieval, software tool calls and exchanges between agents.

Sanjeev Mohan, Principal and Founder of the analyst firm SanjMo, said existing tools were built for that older model of governance and now force a trade-off security teams cannot afford.

"Enterprises have spent years governing data at rest with static policies. AI agents work with data in motion: the chain of thought, retrieved context, tool calls, and handoffs moving in and out of models every second," Mohan said.

"Most inspection tools were built for a pre-AI world and force a bad tradeoff: inspect everything and slow the workload, or sample and hope. Putting deterministic inspection directly in the AI data path is the kind of infrastructure the AI era requires. It is also what frees the most valuable AI use cases, the ones stuck in pilots because teams cannot prove what sensitive data is going into a model or coming back out," he said.

His assessment identifies the blocker Nol8 is selling against: high-value AI use cases that stall in pilot stages because teams cannot demonstrate what data enters a model or returns from it. Deterministic inspection in the data path, in this view, converts an unanswerable compliance question into a checkable rule.

What does deployment look like?

Nol8 says the system ships with operational features aimed at enterprise production environments:

• Availability on AWS for production deployments
• Policy management through a web console or a REST API
• Single sign-on support
• Mutual TLS with automatic certificate rotation
• No storage of customer data and no logging of contents
• A data API for access from existing applications

The no-storage, no-logging claim addresses a circular concern with inspection layers: a tool that watches all AI traffic could itself become the broadest data repository in the company. Nol8 says its plane enforces rules without retaining the material it examines.

What comes next?

The launch positions FortifAI's business unit in a category — AI data governance infrastructure — that barely existed two years ago and that Gartner's and IDC's forecasts suggest will grow as fast as agent adoption itself. Nol8's bet is that deterministic, pattern-based inspection in the AI data path becomes standard plumbing for enterprises that intend to run tens of thousands of agents against their own data, and the benchmark claims of 1 GB/s throughput at 5,000 rules set the performance bar its competitors will now be measured against.