HiddenLayer Pricing: A Quote-Normalization Worksheet

HiddenLayer pricing cannot be reduced to a public per-seat or per-request number from the vendor pages reviewed on 7 November 2026. The platform page names four product areas, while the contact page collects sales inquiries without displaying a rate. A comparable budget starts by separating those product areas and recording the missing public-price fields instead of filling them with guesses.

TL;DR

• HiddenLayer's reviewed public pages show no numeric license price or plan allowance. They also show no billing unit or minimum commitment. No overage rate is displayed.
• The platform spans AI Discovery and AI Supply Chain Security, alongside AI Attack Simulation and AI Runtime Security, which need separate quote lines.
• Use a quote-normalization worksheet for protected assets and scans. Record simulations and routed requests separately. Add deployment and support, then retention and renewal.
• DeepInspect overlaps only where authenticated HTTP model requests are deliberately routed through a runtime enforcement point.

Public-price absence is the first worksheet entry

The current HiddenLayer platform page describes the product surface but displays no numeric license price or plan name in the page content reviewed on 7 November 2026. It also shows no allowance or billing unit. No overage amount or minimum commitment is displayed. The HiddenLayer contact page presents a form with fields for name and email, plus phone number and country or region. It also requests company information but displays no rate card.

Record that result precisely: "No numeric public price found on the reviewed platform and contact pages on 2026-11-07." A contact form proves that the vendor accepts inquiries. It does not prove a particular custom-pricing formula or contract minimum. It also does not prove a discount structure. Those terms require a quote or order form.

I would send back any procurement spreadsheet that converts "contact us" into an invented annual figure. A blank public-price cell is evidence about public disclosure, not permission to estimate the vendor's commercial terms.

HiddenLayer packages several security jobs

HiddenLayer's platform page groups the current product into AI Discovery, AI Supply Chain Security, AI Runtime Security, and AI Attack Simulation. It also names model scanning and red teaming as use cases. AI guardrails and agentic and MCP security are named as well. A single total can therefore conceal different protected assets, operating teams, and usage events.

AI Discovery concerns visibility into AI assets across environments. Supply-chain security covers model integrity before deployment, with model scanning as a named use case. Attack simulation exercises systems to expose weaknesses. Runtime security addresses attacks against production AI. Those jobs create different workload drivers even when the vendor eventually packages them in one commercial proposal.

The model scanning guide explains the artifact boundary: a scanner inspects model files and related packages before deployment. A runtime request control sees traffic after an application invokes a model. Count them separately because ten model artifacts scanned in a registry cannot be compared with ten thousand application requests sent to an LLM.

Build the quote-normalization worksheet by module

Create one worksheet section for each HiddenLayer product area, then add shared commercial terms. Keep the public-price absence at the top so nobody mistakes an internal placeholder for a published rate.

For AI Discovery, capture covered environments and inventoried resource types. Record connectors and discovery frequency. Add duplicate treatment and any charge tied to accounts or subscriptions. Ask how projects and clusters are counted, including any inventory-count charge. Ask how deleted and ephemeral resources affect the meter.

For AI Supply Chain Security, record model repositories and registries. Add file formats, artifact count, and artifact size. Record scan frequency and rescans. Add CI/CD jobs and vulnerability feeds, then retained findings. The quote should identify which scanning capabilities are included and how private deployment changes operations.

For AI Attack Simulation, list the systems tested and campaign count. Record the attack library and scheduled frequency. Add custom scenarios and concurrency. Record human services and report format, plus retest rights. A red-team engagement with analyst time belongs on a different line from continuously scheduled simulation software.

For AI Runtime Security, capture protected applications and model endpoints. Add agents, gateways, and request volume. Record peak throughput and deployment path. Add fail behavior and telemetry export, then retention. The AI security vendor evaluation criteria provide route and identity fields, plus action and evidence fields, for testing the quoted scope.

Normalize shared costs and contract terms

The worksheet needs commercial rows that sit outside the four modules. Record the subscription term and currency. Add the invoicing schedule and implementation services. Record support level and response targets. Add training and connector work. Include private infrastructure and data residency. Record log storage and export, followed by renewal treatment and termination assistance. Put taxes and cloud consumption on their own lines.

Then identify the denominator behind every amount. An annual platform total has little comparison value until the quote states the included assets and scans. It must also state simulations and runtime events, plus environments and support. Preserve the unit exactly as the vendor defines it. "Protected model" may mean a registry artifact, a deployed endpoint, a model version, or something else in the signed terms.

Print the worksheet on one wide sheet if the finance meeting is in person. The physical width is useful: discovery assets sit at the left edge and runtime requests at the right. Shared contract costs run along the bottom. If the numbers only fit after combining those columns, the proposal is still too ambiguous.

Compare cost owners at the correct boundary

Discovery usually attaches to security posture budgets. Scanning can attach to ML supply-chain budgets. Attack simulation may sit with assurance or red teaming, or with application security. Runtime protection can attach to the production AI platform. Model-provider inference and SIEM storage remain separate cost owners. Cloud infrastructure and engineering time also remain separate unless the signed quote explicitly includes them.

The best AI gateway guide separates routing, observability, content screening, and identity-bound policy because products can occupy adjacent positions without sharing a meter. Keep model scans in the artifact count. Attack simulations belong in the campaign ledger. Routed production volume stays with the operating meter.

Only part of HiddenLayer's runtime surface may overlap a DeepInspect deployment. The overlap exists when an authenticated application request to an LLM is deliberately routed through a control point where both products are being considered for runtime policy or protection. Discovery and model-file scanning remain separate jobs. Supply-chain validation and offline attack simulation also remain separate jobs.

Validate the quote with a controlled workload

Before signing, run a bounded test that produces each quoted event type. Add one asset through a supported discovery connector. Scan a known test model artifact twice and learn whether the rescan changes consumption. Run one approved simulation against a disposable target. Send tagged runtime traffic through the proposed integration and reconcile application events with the vendor's count.

Capture screenshots or exports from each surface, plus the quote line that covers it. A good reconciliation identifies the event and timestamp. It also identifies the environment and asset or route. Record the meter name and included allowance. Then record the resulting charge treatment. Repeat the test after a failed request and a retry because retries can expose an unexpected counting rule.

The exclusions also need testing. Ask what happens when a connector fails or a model is replaced. Ask separately about a campaign that stops early or runtime telemetry that arrives late. The signed definition should decide the billable outcome. A salesperson's diagram is useful during discovery, but the order form and service schedule need to carry the counting rule.

DeepInspect

DeepInspect evaluates authenticated HTTP model requests that an application deliberately routes through its proxy. It uses application-supplied identity and role, plus data class and model destination. The policy version is also supplied to make and record a per-request decision.

Price that routed request surface independently. HiddenLayer discovery and model scanning retain their own workload rows. Attack simulation also retains its own workload row, while only the relevant runtime model route belongs in an overlap comparison. Book a technical deep dive at deepinspect.ai.