Use every AI. Surrender none of your data.
AI-DLP Gateway
TrustONE is the data-protection gateway for the AI era. It sits between your people and every large language model they reach for, and makes sure the model only ever sees tokens — never your real data.
The problem
Every AI tool your people adopt is a new way for your data to leave the building.
A prompt to a public model is a transfer you did not design, to infrastructure you do not run, governed by a security team that is not yours. Legacy DLP was built for files, email, and USB sticks. It cannot read meaning, intent, or what a conversation gives away over thirty turns.
Direct disclosure
An analyst pastes the terms of an unclosed deal, source code, or a patient record into a chat window. One prompt, gone.
Accumulated conversation leak
Each message looks harmless. Across a long session the model reconstructs an architecture diagram, a customer list, or a roadmap — piece by piece.
Embedded and agentic AI
SaaS copilots, IDE assistants, and autonomous agents move your data to models silently, without anyone opening a chat window at all.
This is a sovereignty problem, not only a security one.
A prompt sent to a model hosted in another country is a cross-border data transfer. That puts it inside GDPR and Schrems II, the EU AI Act, UAE PDPL, and similar data-residency law — the moment an employee hits enter.
What it is
An enterprise data-loss-prevention platform built for the one channel legacy DLP cannot see: the model itself.

TrustONE intercepts every AI interaction across browser, endpoint, network, and API. It reads the meaning of what is being shared, and enforces your policy in real time — block, mask, or allow — through one engine and one audit log.
It runs standalone, or as the fifth module of the ODUN.ONE platform.
The design principle is simple: blocking AI is impossible. It has to be protected instead.
How it works
Real values stay inside your perimeter. The model only ever sees tokens.
Four steps, every interaction, in under the blink of an eye.
YOUR PERIMETER · data never crosses this line
TrustONE
- 1 · Interceptbrowser · endpoint · network · API
- 2 · Understandsemantic + multi-turn detection
- 3 · Substitutereversible tokenization
The external model is treated as untrusted. Real values never traverse the perimeter boundary — by architecture, not by configuration. A misconfigured policy still cannot leak data.
Capabilities
One platform. Three jobs: protect, control, govern.
Protect
- Semantic and intent detection — understands meaning, not just patterns
- Multi-turn leak detection — catches disclosure accumulated across a conversation
- Reversible tokenization — real values swapped for tokens, then restored
- Multimodal coverage — text, images and OCR, documents, and code
- 100+ languages at the semantic level
Control
- Block, mask, or allow — policy enforced in real time
- Policy-based routing — sensitive work to a private model, general work to public models via tokens
- Token-usage limits by department, user, and model at the same time
- Browser last-mile — covers BYOD and personal AI accounts, before data leaves the device
Govern
- Complete local audit log — every block, mask, and allow recorded, with SIEM export
- Executive dashboards for AI usage and risk
- 300+ LLM services monitored out of the box
- Hardened token vault — AES-256-GCM, per-session ephemeral keys, configurable retention
Why it is different
Four things competitors do not put in writing.
Large vendors now offer self-hosting. These four are where TrustONE still stands apart.
1 True air-gap by default
No control plane, model weight, or audit log ever leaves the customer perimeter. Self-hosting is now common; a genuine air-gap, with no management metadata phoning home, is not.
2 No plaintext ever reaches the model
The downstream model is untrusted. Real values never cross the boundary — enforced by architecture, not by a policy toggle that someone can misconfigure.
3 Reversible, lossless tokenization
Substitution runs both ways. The user gets a fully reconstructed answer, and cardholder or regulated data never transits an external processor — a recognized PCI-DSS scope-reduction practice, confirmed by a customer's QSA.
4 Cumulative multi-turn leak detection
TrustONE detects 88% of leaks that build up across a conversation, against 8% for legacy DLP. This is leak accumulation — a different and harder problem than the jailbreak detection most tools label "multi-turn".
Capability
Sees AI interactions
- TrustONE
- Yes
- Legacy DLP
- No
- Cloud AI gateways
- Yes
Capability
Understands meaning, not just patterns
- TrustONE
- Yes
- Legacy DLP
- No
- Cloud AI gateways
- Partial
Capability
Multi-turn accumulated-leak detection
- TrustONE
- Yes
- Legacy DLP
- No
- Cloud AI gateways
- No
Capability
Reversible tokenization
- TrustONE
- Yes
- Legacy DLP
- No
- Cloud AI gateways
- No
Capability
No plaintext sent to the model
- TrustONE
- Yes
- Legacy DLP
- No
- Cloud AI gateways
- No
Capability
True air-gap deployment
- TrustONE
- Yes
- Legacy DLP
- Varies
- Cloud AI gateways
- No
Competitor marks reflect public vendor materials and are indicative, not bench-certified.
Deployment & sovereignty
Three deployment models. One platform. Your data never leaves your perimeter.
Model A
On-premise
Only tokens reach public models; real data stays on site. Best fit for enterprise, finance, and SaaS. Compliance ceiling: GDPR, HIPAA, PCI-DSS.
Live in 2–3 weeks
Model B
Hybrid
Policy routes sensitive work to a private or local model, general work to public models via tokens. Best fit for pharma, R&D, and banking. Adds CMMC Level 2 and ITAR.
Live in 3–5 weeks
Model C
Air-gap
Zero outbound traffic; all AI runs on-premise on local models. Suited to defense, government, and classified work. Ceiling: CMMC Level 3 and classified.
Live in 4–8 weeks
Channels include Chrome extension, DNS/proxy, endpoint agent (Windows/macOS/Linux), API gateway, SDK, and multi-tenant MSSP mode.
Compliance
When data never leaves, most cross-border compliance questions answer themselves.
Because real values never reach an external model, transfer mechanisms like standard contractual clauses and adequacy decisions stop being the hard part. TrustONE maps to the frameworks that matter:
- GDPR · Schrems II
- EU AI Act
- ISO 27001
- SOC 2
- HIPAA
- PCI-DSS · scope reduction
- CMMC · via air-gap
- FedRAMP
- ITAR / EAR
- UAE PDPL
Sector deployments also map to Basel III, FDA 21 CFR Part 11, EASA Part M, CJIS, and ePrivacy.
Performance
Measured performance, not promises.
- Detection accuracy
- 87–94%
- False-positive rate
- 3–7%
- Multi-turn leak detection
- 88%
- Added latency per prompt (median)
- <50ms
- Availability target (clustered)
- 99.95%
- SOC alert volume vs legacy DLP
- −85%
- LLM services covered
- 300+
- Languages at semantic level
- 100+
| TrustONE | Legacy DLP | |
|---|---|---|
| Detection accuracy | 90% | 52% |
| Multi-turn leak detection | 88% | 8% |
| False positives (lower is better) | 5% | 32% |
A single prevented incident typically pays for the entire platform within the first 60 to 90 days.
Who it is for
Built for organizations where a leak to an AI model is a regulatory, financial, or competitive event.
Industries we protect
- Banking & financial services
- Government & public sector
- Telecommunications
- Pharmaceutical & life sciences
Where TrustONE fits best
Regulated data
Personal, health, or cardholder data under GDPR, HIPAA, or PCI-DSS.
Sovereignty mandates
Rules that forbid data crossing a national or organizational boundary.
High AI adoption
A workforce already using public models daily, faster than policy can keep up.
IP-heavy work
Source code, research, deal terms, or designs that are the business itself.
Blocking AI is impossible. Protecting it is not.
See TrustONE run against your own traffic. A proof of value stands up in weeks, inside your perimeter, with your data never leaving the building.

