Find every model, agent and AI tool in use, including shadow AI. Assess each one, enforce policy on prompts, data and agent actions, and keep the evidence your regulator asks for.
Download the datasheet (PDF)
How Guardrail AI works · Discover. Assess. Govern. Protect. Respond.
01
Discover
Gain complete AI visibility
Discover every model, agent and AI tool, including shadow AI.
02
Assess
Understand risk and impact
Continuously assess risk and business impact.
03
Govern
Establish policy and control
Policies, frameworks and model governance at scale.
04
Protect
Secure AI and sensitive data
Enforce controls, prevent misuse and data leakage.
05
Respond
Detect, investigate and resolve
Detect, investigate and resolve AI incidents fast.
Six stages, from finding an AI system to reporting when it causes harm. Choose a stage to see what each module does.
Risk index, control coverage and open incidents for the whole AI estate in one view.
Risk index, control coverage and open incidents for the whole AI estate in one view.
What sets it apart from a general GRC tool with an AI module.
Built for AI, not adapted to it
Controls are written for models, prompts, agents and tool calls. It is not a general GRC tool with an AI tab added.
Enforces, not just records
Policy runs at the point of use: prompts are inspected, unsanctioned tools are held and agent calls are checked before they execute.
Models and agents in one register
Credit models, GenAI assistants and autonomous agents sit in one inventory, each with an owner, a risk tier and an autonomy level.
Evidence a regulator can follow
Each control links to its evidence, each assessment to its sign-off and each serious incident to the report that was filed.
Visibility
Teams adopt models, assistants and browser AI tools faster than any spreadsheet register can track. Guardrail AI finds them across cloud, SaaS and endpoints, holds unsanctioned tools until they are approved, and gives every asset an owner and a risk tier.
Monitoring
A model approved last year is not the model running today. Guardrail AI monitors fairness and drift on models in production, raises an alert when a threshold is crossed, and links it to the affected EU AI Act and NIST AI RMF controls so the owner knows what to fix.
Audit
Registration, risk assessment, control mapping and evidence collection run as one workflow for each system. When internal audit or the regulator asks, the record is already assembled, with the trail behind it.
Policy packs for AI regulation, controls for the personal data AI uses, and threat catalogs for AI security, each linked to the evidence that shows it is met.
| Framework | Coverage | Status |
|---|---|---|
| EU AI Act | Risk-tier classification, conformity and fundamental rights impact assessments, and serious incident reporting. | Supported |
| NIST AI RMF | Govern, Map, Measure and Manage functions mapped to controls and evidence. | Supported |
| GDPR | DPIAs, records of processing and cross-border transfer checks for AI that uses personal data. | Supported |
| Bangladesh DPA | Personal data controls for AI systems that process data in Bangladesh. | Supported |
| MITRE ATLAS | Adversarial AI techniques, such as prompt injection and data poisoning, mapped to controls. | Supported |
| OWASP Top 10 for LLMs | LLM risks such as sensitive information disclosure and excessive agency, mapped to controls. | Supported |
Regulated sectors where AI already makes or shapes decisions about customers and patients.
Model risk meets GenAI
Banks already validate credit and fraud models, and now run GenAI assistants alongside them. Guardrail AI puts both in one inventory, treats credit scoring as high-risk under the EU AI Act, and keeps assessment and oversight evidence for each.
AI in use
Fair outcomes, on record
Insurers use AI to price risk and triage claims, where an unfair outcome reaches a customer directly. Guardrail AI tests these models for bias, records human oversight of decisions, and keeps an audit trail for each one.
Ship AI features with controls
Fintechs ship AI features quickly, often on third-party models. Guardrail AI tracks which vendors and models each product depends on, controls what data leaves in prompts, and keeps the evidence partners and regulators ask for.
AI near patient data
Healthcare organizations need to know which AI touches patient data. Guardrail AI blocks sensitive data in prompts, checks generated answers against their sources, and tests clinical models for bias.
AI in use
Public AI, answerable
Agencies adopt AI for citizen services and casework under public and oversight scrutiny. Guardrail AI registers every system, records its risk assessment and approval, and logs its decisions so an inquiry can be answered from the record.
AI in use
AI at subscriber scale
Operators run GenAI in customer care and AI across the network. Guardrail AI inventories both, keeps subscriber data out of prompts, and holds the assessment and incident evidence regulators ask for.
AI in use
Working in another regulated sector?
Tell us where AI is in use across your organization and which rules you answer to. We'll show you how Guardrail AI would govern it.
Where Guardrail AI runs, who holds the data and how its AI is kept accountable: the questions your security and procurement teams will ask first.
Deploy your way
Run Guardrail AI on-premises, as SaaS or hybrid. Your team runs it, we co-manage it, or Raptas operates it for you, 24/7.
Your data stays yours
On-premises and hybrid deployments process data inside your infrastructure; SaaS runs in a dedicated environment in your region. Never shared, never used to train models for anyone else.
AI you can account for
Risk scores and classifications show the factors behind them. Unsanctioned tools stay blocked until a person approves them, high-severity fixes need sign-off, and every decision is written to an immutable audit log.
Connects with
An AI governance, risk and compliance product. It finds the models, agents and AI tools in use across your organization, assesses their risk, enforces policy on prompts, data and agent actions, and keeps the evidence auditors and regulators ask for.
Guardrail AI looks for AI use across cloud environments (AWS, Azure, GCP), SaaS platforms and endpoints, including tools such as Hugging Face Spaces, Perplexity AI and Replit AI. Unsanctioned tools are held by default until someone approves them, and approved tools are registered in the inventory.
Policy packs for the EU AI Act and NIST AI RMF, and controls for GDPR and the Bangladesh DPA. For AI security, threats are mapped to MITRE ATLAS and the OWASP Top 10 for LLMs. All are supported, with controls mapped and evidence collected against each.
Guardrail AI monitors live models for group fairness, individual fairness and statistical drift. When a threshold is crossed it raises an alert with its severity and the affected dimensions. High-severity remediation needs human approval.
Yes. It connects to cloud platforms (AWS, Azure, GCP), MLOps platforms (MLflow, Kubeflow), agent frameworks (LangChain, CrewAI, AutoGen), SIEM and SOAR tools, and GRC tools (ServiceNow, Archer, OneTrust). APIs and webhooks cover custom integrations.
On-premises, as SaaS or hybrid. On-premises and hybrid deployments keep prompts, datasets and the AI inventory inside your infrastructure; SaaS runs in a dedicated environment in your region. Your data is never used to train models for anyone else.
Most teams start with the AI inventory and shadow AI discovery, then add assessments, policy and runtime controls as the program matures. Our team handles integration with your environment.
01
Scoping call
A short conversation about your environment, your regulators and what you need to solve first.
02
Demo on your scenario
We walk through Guardrail AI on a scenario you choose, with sample or anonymized data.
03
Pilot in your environment
A time-boxed pilot on-premises, in your cloud or as SaaS, against success criteria agreed up front.
04
Rollout
A phased go-live, with our team on implementation and, if you want it, managed operations.
Security & Operations
Shield Sphere
One SOC console for detection, response and board reporting, in place of the separate tools most teams run today.
Explore Shield SphereSecurity & Operations
Agentic OpsSphere
Ask what's wrong in plain English. Agents work across logs, alerts, changes and topology, then show every step they took.
Explore Agentic OpsSphereRisk & Compliance
GRC Sphere
Move capital, risk and regulatory reporting out of spreadsheets and into a system your auditors can follow.
Explore GRC SphereIn a demo, we take a model or agent like the ones you run and show how Guardrail AI registers it, assesses it against the EU AI Act and NIST AI RMF, enforces policy at runtime and keeps the evidence.