Offerings

Your AI will be attacked.
Let us go first.

NiyamR governs the AI you run, secures every prompt it receives, and attacks your applications, APIs and AI stack before anyone else does. Every finding comes with proof.

Start with a Quick Scan

Quick scan · sandbox

$ scan --mode quick staging.example.com

reconmapping attack surface

exploittesting access control on /api/orders/{id}

validatereproduced with a second account

HighBroken access control (IDOR)

Proof-of-concept attached. Reproduction steps included.

reportwrite-up and SARIF ready

Illustrative run

PoC
on every reported finding: 100+
AI components fingerprinted: 42
enforceable policy categories: 0
prompts sent to a third-party model

Why now

AI is everywhere. Control is not.

Blocking AI just pushes it out of sight. Make safe use the easy path.

92%
of firms whose AI was attacked had no proper AI access controls: 39.7%
of all AI interactions at work involve sensitive data: 47%
of AI users at work still use personal AI apps nobody monitors: €35M
or 7% of global turnover: the top fine under the EU AI Act

Quick Scan · The fast way in

Not ready for a full engagement? Get an answer in minutes.

Point us at one target. A Quick Scan runs the same agents as a full test, with a shorter budget. No long scoping call, no procurement cycle.

01Minutes

Quick Pentest Scan

You give: A URL, a repository or an API spec
You get: Validated findings, each with a proof-of-concept

02Minutes

Quick AI Exposure Scan

You give: A domain or an IP range
You get: The AI components you expose, and their known CVEs

03Seconds

Quick Prompt Check

You give: A prompt, pasted in your browser
You get: What Shield would allow, redact or block

Try it live

A Quick Scan is a first look, not a full penetration test. It covers what it analysed in the time it had. Authorized targets only.

One platform · One loop

Decide the rules once. Enforced everywhere.

01SeeFind every AI in useGovernance
02AssessScore it against the lawGovernance
03DecideWrite the rules onceGovernance
04TestAttack it on purposePentest · Red Team
05ProtectEnforce rules on live trafficSecurity
06ProveKeep every action as evidenceAll

The parts feed each other. A finding from a test becomes a task with an owner, AI Security blocks the gap on live traffic, and every action is kept as evidence.

AI Governance

Can you prove your AI complies?

One record of every AI system, assessed against the regulation.

Find the AI in your organisation, assess it against the regulation text, and keep the policy and the evidence in one place.

Frameworks: EU AI Act · UK · MAS · GDPR · NIST RMF
Coverage: Sanctioned and shadow AI
Output: Cited gaps and evidence

Explore AI Governance

Compliance Assessment

Each AI system is assessed against the regulation text, not a checklist.

Cited to the article
Multi-jurisdiction
Evidence collected as you go

You leave withA gap analysis a regulator can follow

Shadow AI Discovery

Finds the AI tools and integrations nobody registered.

Gateway, browser and network signals
Microsoft 365 and Google Workspace
Risk scored and ranked

You leave withA complete AI asset registry

Policy Tracker

One policy set, mapped to every jurisdiction you operate in.

Obligations mapped to articles
Live block, mask or log rules
Drift detection

You leave withOne rule set, enforced everywhere

Evidence and Audit Trail

A record of what was assessed, decided and enforced, ready for a reviewer.

Every gap becomes a task with an owner
Metadata only, no prompt text
Roles for admins, operators and viewers

You leave withReports ready for audit

Compliance Copilot

Ask in plain English. Answers come from your own data and the regulation text.

Grounded in your AI register
Cites the regulation

You leave withAnswers your team can act on

AI Security

What leaves your company in a prompt?

Enforce your policy on every prompt, response and agent.

Controls that sit in the path of AI traffic. Detection runs on your own hardware, so prompt text never reaches a third-party model.

Actions: Allow · Redact · Block · Log
Detection: Runs locally
Policy: 42 categories

Explore AI Security

NiyamR Shield for Browser

Checks every prompt, paste and file attachment before it leaves the user’s machine.

17 AI products
Secrets, PII, PHI, your own terms
Redact with Undo

You leave withEmployees keep using AI, without the leak

Shield Gateway

An inline proxy for the AI API traffic your applications send. Prompts and replies are both checked.

OpenAI, Anthropic, Azure OpenAI compatible
Injection and jailbreak detection
Scan API for systems that cannot reroute
Runs in your own cloud

You leave withOne enforcement point for every AI call

Contained Agent Security

Shows what your agents and MCP servers can be made to do, and keeps it contained.

14 MCP risk categories
Tool poisoning
Excessive permissions

You leave withA risk map of every agent and tool

Where it protectsThree levels, stated plainly

Stopped in real time: AI your apps call in code · AI websites on work browsers. Every prompt and reply is read.
Seen and governed: Desktop and phone AI apps · AI built into SaaS tools. Allowed or denied by policy.
Policy only: Personal devices. Covered by acceptable-use policy and training. We do not claim more than that.

Penetration Testing

Would you find it before an attacker does?

An autonomous agent that tests your whole application like a real attacker.

Not only AI. A team of agents tests your web apps, APIs, code and cloud, exploits what it finds inside a sandbox, and reports only what it can prove. Pick the depth: quick, standard or deep.

Depth: Quick · Standard · Deep
Targets: Code · Web apps · APIs · Cloud
Findings: Validated with PoC

Explore Penetration Testing

Quick Scan

Minutes

Fast checks on one target. The same agents as a deep test, on a shorter budget.

A URL, repository or API spec
Highest-impact checks first
No long scoping call

You leave withValidated findings, the same day

Standard Assessment

Balanced

The balanced option for routine testing. Broader than a quick scan, faster than a deep test.

Full vulnerability coverage
Live URL, code or both
Fix guidance per finding

You leave withA routine assessment you can repeat every release

Deep Penetration Test

Hours

Full-depth recon, exploitation and validation. Agents share discoveries and chain vulnerabilities.

White, black or grey-box
OWASP Top 10 and business logic
Cloud and infrastructure
Generated patches

You leave withA full pentest report with a PoC per finding

Continuous Security Gating

Every pull request

The same tests on every pull request, scoped to the changed files.

Exit codes for pipeline gating
Results in your code-scanning tool

You leave withNo release ships untested

What it testsOWASP Top 10 and beyond

Access control: IDOR · Privilege escalation · Broken authorization
Injection: SQL · NoSQL · OS command · SSTI
Server-side: SSRF · XXE · RCE · Deserialization · File uploads
Client-side: XSS · CSRF · Prototype pollution · Open redirect
Authentication: JWT · OAuth · Sessions · Weak passwords
Business logic: Race conditions · Payment abuse · Mass assignment
Cloud and infrastructure: AWS · Azure · GCP · Kubernetes · Dependency CVEs
AI-specific: Prompt injection · Agentic system security

AI Red Teaming

How does your AI hold up under attack?

Adversarial testing for models, agents and the stack around them.

Conventional scanners do not cover model servers, LLM gateways or agents. This does, and it validates a finding before it reports it.

Techniques: 196+ · OWASP LLM Top 10
Scope: Models · Agents · AI stack
Closes with: A re-test

Explore AI Red Teaming

AI Infrastructure Scan

Quick Scan available

Finds the exposed and vulnerable parts of your AI stack.

100+ AI components
Known-CVE matching
Validated before reporting

You leave withConfirmed exposures, not version guesses

LLM Jailbreak Evaluation

Measures how your model holds up under sustained attack.

Single and multi-turn attacks
PAIR, GOAT, Crescendo and more

You leave withA success score per attack method

Red Team Readiness Pack

The full assessment as one scoped engagement, with one evidence pack.

Executive and technical report
CVSS and SARIF 2.1.0
Verification re-test

You leave withEverything a buyer or auditor asks to see

What we attackOne AI system at a time

Prompt injection: Hidden orders slipped into text
Jailbreaks: Tricks that make it break its rules
Data leakage: Getting it to reveal private data
Agent hijacking: Taking over its tools and actions
System prompt leaks: Exposing its instructions
Harmful output: Unfair, unsafe or false answers

Why NiyamR

The depth of a manual test. The speed of a scanner.

	Manual pentest	Static scanner	NiyamR
Time to result	Weeks	Minutes	Minutes to hours
Findings	Validated	Unvalidated, noisy	Validated, with PoC
How often	Once or twice a year	Every build	Every change
AI stack coverage	Depends on the tester	Rarely	Models, MCP, agents
Fix verification	A new engagement	Re-scan	Re-test included

Who it is for

Your sector. Your sensitive data.

Banks and finance

Client names
Account numbers
Source-of-wealth notes

Healthcare

Patient names
Case numbers
Diagnoses

Pharma and CROs

Compound IDs
Unpublished trial results
Sponsor material

Law firms

Party names
Matter references
Privileged text

HR and recruiting

Candidate names
Home addresses
Salary history

Retail and travel

Customer names
Card numbers
Booking references

Where to start

Tell us who you are. We show you the path.

Stop the leak, then test the surface.

01Quick ScanSee real findings today.See details
02NiyamR Shield for BrowserClose the prompt leak.See details
03Deep Penetration TestGet the full test, with proof.See details

Red teaming runs only against systems you own or have written permission to test. A clean result covers what was in scope, not what was left out.

How we get started

Start small. Prove it. Then roll out.

Phase 1
Browser pilot
The extension runs in watch-only mode. See what it would have caught.
Phase 2
Compliance pack
Your AI register built, scored against the rules, and ready for audit.
Phase 3
Security test
One system attacked, fixed and re-tested, with a shareable report.
Phase 4
Full rollout
Gateway, network feed and one policy enforced across the company.

Privacy by design

Your data stays yours.

A security tool that leaked your prompts would defeat its purpose. Ours is built so it cannot.

Backed by Innovate UK · NVIDIA Inception

No prompt text kept: We keep the verdict, category, time and incident ID. Never the text.
Scanned in memory: Checked in memory, then thrown away.
No third-party AI: Shield judges with our own models, never another AI firm.
Keys never stored: Provider keys pass through. They are never saved or logged.
Your region: Scanning runs where you choose. EU data stays in the EU.
Or in your cloud: For strict cases, the gateway runs in your own cloud.

Ready to institutionalize AI safety?

Start with the registry. Every rule runs in monitor mode on your own hardware until you have seen your own traffic and decided what to turn up.

Deploy Registry Free Contact Engineering

Data stays in your tenant

Prompts are not retained

DPA available on request

Sub-processors published

Your AI will be attacked.Let us go first.