Mask sensitive prompt data before it leaves AI apps.
NeutralAI adds a compliance-first control layer for browser AI and app traffic, with masking, encrypted tokenization, and audit-ready proof for regulated teams.
Auto evidence
SOC2 readiness / GDPR-aligned / Cyber Essentials via review
Token vault
AES-256-GCM
Measured overhead
~41 ms
Live Narrative
How NeutralAI works in one view
Raw input
Sanitized prompt
Audit evidence recorded
3 entities tokenized · policy PII-Mask-v3 · export-ready
How NeutralAI Protects Prompts
Intercept
Prompt traffic is captured before external model routing.
Detect
Policy-aware recognition flags sensitive identifiers early.
Mask
Sensitive values are replaced with safer tokens or sanitized references.
Audit
Evidence and policy decisions stay review-ready for security teams.
Why teams choose it
Industries
Use cases buyers can recognize.
Industry pages help regulated teams map NeutralAI to the workflows, identifiers, and review questions they already face.
Financial services
Regulated workflow
Mask customer, payment, account, and claim identifiers before AI prompt egress.
View use caseHealthcare
PHI-aware controls
Protect PHI-style identifiers before clinical, support, or operational prompts reach AI providers.
View use caseLegal
Matter confidentiality
Keep client, matter, and document-review context under a controlled AI gateway path.
View use caseAdoption Without Friction
Secure AI usage without changing habits
NeutralAI works best when adoption does not require a behaviour reset. Teams keep familiar browser-based AI tools while NeutralAI adds prompt protection, auth context, and policy support in the background.
Same tab. Same prompt box. Protected underneath.
NeutralAI can protect browser-based AI usage in the flow people already know, which is exactly why adoption can move faster.
User Experience
People keep the workflow. NeutralAI adds the protection layer.
What Users Feel
No workflow disruption, no extra friction, and no new daily habit to learn.
What Security Gets
A real control point over browser-based AI usage instead of hoping people self-police prompts.
Deployment Options
One product, multiple deployment paths
NeutralAI is not a single hosting story. Teams can choose the operating model that fits their risk posture, infrastructure constraints, and rollout speed.
SaaS
Fastest path for teams that want managed rollout
Use NeutralAI as a managed service when you want speed, lower operational overhead, and a simpler path into protected AI usage.
Private Cloud
For teams that need stronger environment control
Run NeutralAI in a customer-controlled cloud environment when governance, network boundaries, or data posture require more separation.
On-Prem
For regulated deployments with strict infrastructure requirements
Deploy NeutralAI inside your own infrastructure when policy, compliance, or customer obligations demand the highest level of control.
Why It Matters
AI usage grows faster than approval paths.
The real problem is not just privacy. It is losing the ability to approve AI usage with confidence before shadow workflows become normal.
Legal buyer signal
Law firms get blocked when confidentiality concerns, breach-response duties, and client trust reviews arrive before a safe AI control path exists.
See legal AI use caseSensitive data leaves first
Client names, matter IDs, claim references, transcripts, and internal context can leave the boundary before policy is applied.
Security becomes the blocker
Without a visible control point, legal and security teams struggle to answer confidentiality, breach, and insurer review questions.
Shadow AI becomes normal
When approved workflows lag behind demand, people improvise with unmanaged tools and client trust erodes before procurement can respond.
Live Preview — No Signup
See masking in action
Type or paste any text containing sensitive data below. Watch PII get masked in real time — no account, no network call.
Masked Output
5 entities maskedDetected & masked
- PERSON“John Doe”→
<PERSON> - EMAIL“[email protected]”→
<EMAIL> - CREDIT_CARD“4111 1111 1111 1111”→
<CREDIT_CARD> - NI_NUMBER“AB 12 34 56 C”→
<NI_NUMBER> - PHONE“+44 7700 900 123”→
<PHONE>
Live preview uses client-side regex heuristics only. The gateway uses Presidio + NER for production-grade detection. Try the full playground for server-side accuracy.
How It Works
Three steps. One control layer.
NeutralAI sits between the workflow and the model, intercepting traffic before raw sensitive values continue downstream.
Intercept
Traffic hits NeutralAI before it reaches the external model.
Neutralize
Sensitive values become safer tokens or sanitized references.
Forward
Only the cleaned request continues downstream.
Detection Engine
The technical detail buyers ask for, without the wall of docs.
NeutralAI combines real-time recognizers, semantic validation, tenant rules, and masking mode controls so security teams can understand what happens before prompt egress.
Stage 1
Presidio NER + Pattern Matching
Real-time recognizers catch common personal, financial, regional, and network identifiers before the request moves downstream.
Stage 2
Semantic Validation via Qdrant vector DB
Context checks help reduce false positives before policy decisions, with confidence thresholds configurable per entity type.
Entity Types Grid
Visible coverage for common PII classes
Benchmark Coverage
Current multilingual benchmark scope covers 10 target languages; additional packs should be promoted after approved benchmark releases.
Masking Modes
Irreversible
<EMAIL>Sensitive value is removed from the prompt path.
Reversible
<EMAIL_token_abc123>Token is stored in an encrypted vault and retrievable only with authorization.
Two-stage detection
Presidio NER and pattern matching with semantic validation using Qdrant.
Encrypted token vault
AES-256-GCM-backed reversible tokenization for governed restore paths.
Entity coverage
EMAIL, PHONE_NUMBER, PERSON, CREDIT_CARD, IBAN, SSN, TR_ID_NUMBER, UK_NHS_NUMBER, and custom rules.
Policy tuning
Configurable confidence thresholds per entity type and multilingual detection coverage.
Why Trust NeutralAI
Proof your security team can trust.
NeutralAI goes beyond masking by combining policy enforcement, encrypted tokenization, audit-ready evidence, and deployment options built for regulated AI adoption.
Compliance evidence automation
NeutralAI gives teams a control point that can produce audit-ready proof instead of leaving AI usage invisible.
Reversible vault tokenization
Sensitive values can be replaced with encrypted tokens before model egress, then restored only through governed paths.
Managed now, private cloud/on-prem planning
Teams can start with managed SaaS and move toward stricter deployment models as governance demands increase.
Benchmark proof
Measured against a reproducible Presidio-vanilla baseline.
NeutralAI combines proven open-source detection primitives with multilingual calibration, masking, and enforcement layers. The gateway repo remains the measurement source of truth, while the website links buyers to the published methodology and benchmark surface.
Product benchmark, not a third-party independent evaluation.
Public overall F1
99.8%
Holdout overall F1
98.4%
Holdout PERSON F1
92.7%
Healthcare trust
HIPAA-ready deployment support without blanket claims.
NeutralAI helps healthcare and healthtech teams protect PHI before prompts reach AI providers, with PHI-aware controls, audit evidence, and BAA review support for eligible deployments.
Not legal advice. BAA terms, deployment model, and customer obligations require review.
PHI-aware masking
Patient names, contact details, medical record numbers, health plan/member IDs, and device/UDI-style identifiers can be handled before model routing.
Minimum-necessary posture
Prompts keep useful clinical or operational context while direct identifiers are reduced before they leave the approved workflow.
Review-ready evidence
Audit metadata, breach workflow support, and an evidence pack are available under review/NDA without putting raw PHI into standard reports.
BAA review support
BAA review is available for eligible healthcare deployments, with final terms and deployment responsibilities reviewed commercially.
Document proof
Protect files before document content reaches AI workflows.
NeutralAI document handling extends PII protection beyond chat prompts, with document-aware extraction, redaction output, and audit-safe finding metadata for supported upload flows.
Supports simple text PDFs today. OCR-backed image detection depends on configured OCR runtime.
PDF redaction
Supports simple text PDFs today and returns generated PDF output with visual blackout marks.
Office and images
Extracts Office document text and supports OCR-backed image detection when OCR is configured.
Audit-safe metadata
Records file hash, page count, finding counts, and approximate locations without raw sensitive text in standard logs.
Operational Signals
Live product. Easy to verify.
You can quickly check that NeutralAI is live: api.neutralai.co.uk is public, benchmark pages are published, and gateway latency is tracked separately from model response time.
Pricing
Pricing is being finalised
We are finalising our published pricing ahead of commercial launch. Plans start free, and paid tiers are sized for small and mid-sized regulated teams — talk to us for current plans and early-access pricing.
Start in minutes
Put the control layer in place before the rollout gets messy
NeutralAI is for teams that already know AI usage is happening and want a credible way to reduce prompt risk without slowing everyone down.
Need a security or commercial conversation first? Contact NeutralAI.
Social Proof
Proof buyers can verify.
Evidence for regulated teams: supported industries, public benchmark scope, measured gateway overhead, and evaluation patterns without invented customer claims.
Proof-backed metrics
Concrete signals buyers can verify.
These numbers come from product surfaces, benchmark artifacts, and documented website claims. Production usage counts and customer outcomes are published only when an approved source exists.
20+
PII entity types
Names, contacts, account identifiers, cards, IBANs, NHS-style IDs, and custom rules.
10
benchmark languages
Current multilingual benchmark scope across English, Turkish, German, French, Spanish, and more.
99.8%
public overall F1
Gateway-owned product benchmark, not a third-party independent evaluation.
~41 ms
measured overhead
NeutralAI gateway overhead measured separately from model generation time.
Finance evaluation pattern
A regulated team wants AI summaries without leaking customer identifiers.
NeutralAI sits in front of prompt traffic, masks payment and contact details, and gives security reviewers evidence before wider rollout.
Healthcare evaluation pattern
A healthtech team needs useful AI output while reducing raw PHI exposure.
Prompts keep operational context while direct patient identifiers are tokenized or removed before external model routing.