Lakera vs HiddenLayer

Lakera and HiddenLayer side by side: what each one is, what it costs to start, and whether you can try it free. Everything below comes from the vendors' own sites and other public sources, not from hands-on testing.

At a glance

Prices are in US dollars unless shown otherwise. Confirm the current price on the vendor's site before you buy.

Key differences

Lakera lists a free tier and HiddenLayer does not.

Pricing is not directly comparable from the numbers we hold (Lakera: Free; HiddenLayer: Custom pricing), so check each vendor's pricing page.

Both are in our Cybersecurity AI category.

About each tool

L

Lakera

Runtime guardrails for AI apps and agents, sold by Check Point. Community tier up to 10k requests a month; Enterprise pricing is not published.

Who it is for

Security teams and organizations building or deploying LLM apps and AI agents, according to the vendor's docs. The Community tier is meant for evaluation at low volume, while larger volumes, self-hosting and extra controls are listed for Enterprise.

Strengths

• The docs say AI Guardrails works with any model or provider, including hosted (OpenAI, Anthropic, Cohere), open-source and custom or fine-tuned models.
• The docs say AI Agent Security discovers agents, their tools and connected MCP servers, rates risk per agent, and applies guardrails to prompts, outputs, tool calls, tool responses and tool descriptions.
• The data leakage docs list built-in detectors (names, US addresses, US phone numbers, emails, IP addresses, credit card numbers, IBANs, US Social Security numbers) plus custom guardrails, trigger words and custom entity types, with a choice to block or mask.
• Check Point's privacy data sheet lists customer-selected hosting in the EU, U.S. or Singapore for the cloud service.
• Self-hosting is documented via Helm on Kubernetes, Docker or another OCI runtime, or air-gapped; the docs say it is meant to keep all data in the customer's infrastructure.
• The docs say Community accounts get 10k screening requests a month at no stated charge, which allows testing before talking to sales.

Limitations

• Community is limited to 10k screening requests a month, per the official docs. The same page lists 1MB of context per request, RBAC for the Dashboard, SIEM export to an S3-compatible bucket and data retention control as Enterprise features.
• Self-hosting requires an AI Guardrails Enterprise license, and the docs say the Dashboard is not currently available to self-hosting customers.
• No prices are published. A June 2026 third-party review says Lakera does not publish list prices and that new paid quotes go through Check Point enterprise sales (Reported).
• Check Point's data sheet says cloud data is retained for the subscription plus three months after termination; customers can delete prompts or restrict retention of pseudonymized prompts. The data sheet does not address training on customer data, though the data leakage docs say Check Point's AI models will not be trained on PII.
• The vendor's docs do not describe the detection techniques or models behind the guardrails. Earlier third-party write-ups mention an 8,000-token prompt limit on Community and a possible Pro tier; neither could be confirmed on an official page (Reported).

Lakera profileVisit site ↗

H

HiddenLayer

AI security platform to inventory AI assets, scan models, simulate attacks and monitor runtime; pricing is not published and is by demo request.

Who it is for

Enterprise security and ML teams that build or run AI models and agents and want to inventory them, scan models before deployment, test them with simulated attacks and monitor them in production. The vendor names financial services, technology, US federal and healthcare as target sectors, and no self-serve or individual plan is shown.

Strengths

• The AWS Marketplace listing says AI Discovery builds an inventory of agents, models, endpoints and workflows, including shadow AI.
• The same listing says AI Supply Chain Security scans model artifacts for malware, backdoors and vulnerable dependencies and generates an AI Bill of Materials (AIBOM).
• HiddenLayer lists continuous adversarial testing aligned to MITRE ATLAS (System Prompt and Red Team evaluations) and runtime policies that can Block, Detect or Redact.
• The AWS listing names integrations with GitHub Actions, Splunk, Microsoft Sentinel, JFrog Artifactory, Databricks Unity Catalog and Hugging Face gated repositories, and coverage of Amazon Bedrock and SageMaker. The vendor site also lists AWS and Databricks as partners.
• The vendor says it holds SOC 2 Type 2 (renewed with five Trust Services Criteria) and ISO 27001, per a post dated February 13, 2025, and its Security page says customer data stores are encrypted at rest and data in transit uses TLS 1.2 or higher.
• The CrowdStrike Marketplace listing says the platform can monitor AI systems without access to model code or training data and send AI threat intelligence to the CrowdStrike console.

Limitations

• No pricing, free plan or free trial is published on the vendor site or the CrowdStrike listing; the offer is a demo request.
• The AWS Marketplace listing shows a single 12-month contract at $5,000,000 sold in Units that the page does not define; it says all fees are non-refundable and non-cancellable except as required by law, and AWS infrastructure costs may apply.
• The Security page lists a data retention and disposal heading but gives no retention periods, and neither it nor the privacy notice says whether customer data is used to train models; the privacy notice covers website visitors and sales contacts, not product data.
• Vendor support on the AWS listing is by email, Monday through Friday.
• The vendor pages do not explain how AI or ML is used inside HiddenLayer's own detection and scanning, beyond the Security page naming a Machine Learning Detection & Response System.

HiddenLayer profileVisit site ↗

All comparisons · Browse the full directory →