NativeAI Guard
Complete Technical Reference
Swiss sovereign AI data loss prevention · Real-time enforcement · Anonymization engine · Semantic DLP · On-premises or Swiss SaaS
NativeAI Guard is a Swiss sovereign data loss prevention (DLP) and control layer for enterprise AI use. It inspects every prompt, upload, and AI agent interaction in real time, masks, blocks, or coaches before sensitive data leaves the organization. All processing runs on NativeAI's own servers in a Swiss datacenter in Lausanne, under Swiss law, with no public cloud and no exposure to the US CLOUD Act.
01 · At-a-glance specifications
| Product | NativeAI Guard. Two modules: NativeAI Guard Employee DLP (browser DLP) + NativeAI Guard Agent DLP & Firewall (API proxy). |
|---|---|
| Vendor | NativeAI Sàrl. EPFL Innovation Park, Bâtiment C, 1015 Lausanne, Switzerland. |
| Data residency | Switzerland only. No public cloud. No AWS, Azure, or GCP. No US CLOUD Act exposure. |
| Enforcement | Browser extension (NativeAI Guard Employee DLP) + drop-in API proxy (NativeAI Guard Agent DLP & Firewall) |
| Browsers | Chrome, Microsoft Edge (production). Firefox, Safari (roadmap, developed based on client demand). |
| LLM coverage | Provider-agnostic. Any web-based AI tool covered at the browser layer. Named: ChatGPT, Copilot, Claude, Gemini, Grok, DeepSeek, Perplexity, DeepL, and 200+ surfaces. |
| Actions | Anonymize (replace sensitive fields, keep working), Block (with explanation), Coach (in-flow notification), Allow (log only) |
| Detection | GPU-accelerated, LLM-based semantic detection. PII, client-identifying data, patient data, secrets, confidential business information. Beyond regex: understands context and intent. |
| Policy engine | Natural-language rules. Standard base policies (DSG, GDPR, banking secrecy, insurance secrecy, medical secrecy) managed and updated by NativeAI as regulations evolve, included in the license. Custom policies created during onboarding based on client-specific data structures. Client can fine-tune and optimize policies independently. Policies assignable per team, department, or user group. |
| Encryption | TLS 1.3 in transit, AES-256-GCM at rest |
| Authentication | OIDC (PKCE) or SAML 2.0. Connects to customer IdP (e.g. Entra ID / Azure AD). MFA enforced via IdP. |
| Tenant isolation | Dedicated Kubernetes namespace + separate PostgreSQL instance per customer |
| Compliance | DSG (nFADP), GDPR, EU AI Act, FINMA Guidance 08/2024, FINMA Circular 2023/1, FINMA Circular 2018/3, FINMA SC 05/2020, DORA, NIS2, Art. 47 BankG, Art. 321 CP, Art. 35 VAG, Art. 84 KVG |
| Security posture | Aligned with ISO 27001; certification in progress |
| SIEM | REST API, WebSocket real-time streaming, structured JSON events. Syslog (RFC 5424) and CEF on roadmap. |
| Deployment | Swiss SaaS (own servers, Lausanne) or fully on-premises (Kubernetes, signed OCI images) |
| Pricing | NativeAI Guard Employee DLP: CHF 4 to CHF 16 per user/month depending on volume (flat rate for SME). NativeAI Guard Agent DLP & Firewall: metered, usage-based, pricing on request. No implementation fees. |
| Pilot phase | Fully managed, from CHF 5,000 (30 users, 1 month). 100% credited to Year 1 subscription. |
· v1.2 (demo invitation) · v1.1 (product names): 2026-09-07 · v1.0: 2026-08-15
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 1
NativeAI GuardContents
Contents
This reference documents the architecture, deployment, compliance alignment, integration surface, and commercial terms of NativeAI Guard. Page numbers refer to the printed document.
Document scope
This is a product reference, not a contract. Commercial terms stated here are indicative and are confirmed in the pilot engagement letter or subscription agreement. Regulatory mappings describe how NativeAI Guard contributes evidence toward the named obligations; they are not a legal opinion on compliance.
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 2
NativeAI GuardArchitecture
02 · Enforcement architecture
Four enforcement components
| Browser Extension | Production-ready (Chrome, Edge). Intercepts outbound content (prompts, emails, file uploads, web forms) in the browser before it leaves the endpoint. Provider-agnostic: any web-based AI tool is automatically covered. |
|---|---|
| Policy Engine | Production-ready. GPU-accelerated content analysis. Processes content transiently in GPU memory (milliseconds). Detects PII, evaluates natural-language policies, anonymizes where needed. |
| Desktop Agent | In development (Windows). Extends protection to desktop applications (ChatGPT desktop app, Claude desktop app). Enforcement path validated in the conformance lab. Additional platform support developed based on pilot requirements and client demand. |
| API Proxy | Production-ready. Drop-in proxy between internal systems (agents, RAG pipelines, chatbots, voice agents) and LLM APIs. Primary component for NativeAI Guard Agent DLP & Firewall. |
How enforcement works
NativeAI Guard Employee DLP
The Browser Extension intercepts content before it leaves the browser. Content is sent to the Policy Engine, which analyses it transiently in GPU memory. The engine decides: Allow, Block, or Anonymize. The extension enforces the decision. End-to-end latency: milliseconds.
NativeAI Guard Agent DLP & Firewall
The API Proxy sits between your internal systems and the LLM API. Every prompt is inspected on the way in (prompt injection detection, jailbreak detection). Every response is inspected on the way out (data exfiltration prevention, toxic output filtering). No changes to your agent code beyond pointing the API endpoint at the proxy.
Data handling model
By default, the Policy Engine processes content transiently in GPU memory with zero persistence. For industries where compliance requires a full record of AI interactions, NativeAI Guard offers optional persistent storage of prompt content and AI responses. The customer controls which mode applies.
| Transient (default) | Content processed exclusively in volatile GPU memory. Memory actively nulled after analysis. No writing to disk, database, logs, or cache. |
|---|---|
| Audit metadata (always) | Policy decision, timestamp, user ID, target domain, enforcement action, triggered policy rule. No prompt content stored in this layer. |
| Persistent (optional) | Full prompt content and AI responses stored, encrypted at rest (AES-256-GCM). Customer-controlled retention period. Enables full audit trail and forensic reconstruction. Available for compliance-driven industries. |
Access control (RBAC)
| Admin | Full access: policy configuration, user management, audit logs, system settings |
|---|---|
| Analyst | Audit log access, incident investigation, policy review. No administrative changes. |
| Viewer | Read-only access to dashboards and audit logs |
Authentication: OIDC (PKCE) or SAML 2.0, connected to the customer's existing identity provider (e.g. Entra ID / Azure AD). MFA enforced via IdP.
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 3
NativeAI GuardPolicies & Security
03 · Team-based policy assignment
Compliance policies are not one-size-fits-all. Different departments handle different types of sensitive data and face different regulatory obligations. NativeAI Guard allows administrators to define distinct policy sets and assign them to specific teams, departments, or user groups.
| Group-level policies | Define separate compliance policies per team, department, business unit, or user group. Each group can have its own detection rules, enforcement actions, and allowed AI tools. |
|---|---|
| Example: claims | An insurance company's claims team handles medical data (Art. 321 CP, Art. 84 KVG). Their policy blocks all patient identifiers, diagnosis codes, and treatment details. Enforcement action: Block. |
| Example: HR | HR handles employee PII, salary data, and performance reviews. Their policy detects and anonymizes employee names, compensation figures, and performance assessments. Enforcement action: Anonymize. |
| Example: legal | Legal handles client-privileged communications and M&A data. Their policy blocks content matching privileged communication patterns or deal-related terminology. Enforcement action: Block. |
| IdP group sync | User groups sync from the customer's identity provider (Entra ID / Azure AD). When an employee moves departments in the IdP, their policy assignment updates automatically. |
| Policy inheritance | Organization-wide baseline policies apply to all users. Team-specific policies layer on top, adding stricter rules where needed. No team can be less restrictive than the baseline. |
04 · Policy lifecycle
Standard base policies
NativeAI provides and manages base compliance policies covering DSG, GDPR, banking secrecy, insurance secrecy, medical secrecy, and other regulatory frameworks. These are tested, optimized, and updated by NativeAI as regulations change. Included in the license, no additional cost.
Custom policies + self-service
During onboarding, NativeAI works with the client's team to create custom policies tailored to their data structures, document types, and use cases. Included in the pilot phase. After onboarding, the client can independently fine-tune and optimize policies via the admin console.
05 · Platform support
| Chrome (extension) | Production-ready. Primary supported browser. |
|---|---|
| Edge (extension) | Production-ready. Chromium-based; fully supported. |
| Firefox (extension) | Roadmap. Developed based on client demand. |
| Safari (extension) | Roadmap. Developed based on client demand. |
| Windows Desktop Agent | In development. Extends protection to native desktop apps (ChatGPT, Claude desktop). |
| macOS Desktop Agent | Roadmap. Developed based on pilot requirements and client demand. |
| Linux Desktop Agent | Roadmap. Developed based on pilot requirements and client demand. |
The Browser Extension covers all operating systems (macOS, Linux, Windows) independently. The Desktop Agent is not required for a pilot phase.
06 · LLM providers covered
NativeAI Guard is provider-agnostic. The Browser Extension operates at the browser layer, which means any web-based LLM service is automatically covered. Named providers currently used by enterprise customers:
ChatGPTMicrosoft CopilotClaudeGoogle GeminiGrokDeepSeekPerplexityDeepLMistral Le ChatMeta AIHuggingChatCohere CoralGroq ChatTogether AIOpenRouter
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 4
NativeAI GuardProtected Channels
07 · Which AI tools and web surfaces does NativeAI Guard protect?
NativeAI Guard operates at the browser layer, so every web-based surface where an employee can type, paste, or upload data is protected automatically. The tools named below are examples, not a complete list. Coverage is not restricted to these products: any comparable web service is covered by the same rules, and the list grows with pilot requirements and client demand.
AI chatbots and LLMs
AI code assistants
Translation services
AI writing and productivity
Search engines with AI
AI image generators
Social media and messaging (web)
Email (web-based)
Cloud storage and documents
AI meeting and note tools
Enterprise SaaS AI features
AI video and audio tools
AI research and data analysis
Other exfiltration surfaces
The categories above are illustrative. Any web surface that accepts typed, pasted, or uploaded content falls under the same policy engine, whether or not the product is named here.
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 5
NativeAI GuardDeployment & Compliance
08 · Deployment
Where is data processed and stored?
Swiss SaaS
NativeAI's own servers in Lausanne
Swiss company, Swiss law, Swiss engineers. No public cloud, no AWS, Azure, or GCP anywhere in the stack. Data that never reaches a US provider cannot be compelled under the US CLOUD Act or FISA 702. Zero-downtime updates deployed continuously.
On-premises
Customer's own infrastructure
NativeAI Guard deploys fully inside the customer's own infrastructure as signed OCI container images on Kubernetes. All data processing stays within the customer's perimeter. Source code is never shared.
On-premises requirements
| Container runtime | Kubernetes v1.25+ with OCI image support |
|---|---|
| GPU | NVIDIA H100 or H200 GPU nodes for the Policy Engine. Multiple GPU nodes required depending on user count and throughput. This is the primary cost driver for on-premises deployments: customers who do not already operate GPU infrastructure should factor in the hardware investment. |
| Storage | Persistent storage for audit logs (PostgreSQL). Capacity depends on retention period and user count. |
| Network | Outbound connectivity for browser extension communication. All data stays in-perimeter. |
| Security | Restricted Pod Security Standards, default-deny Cilium NetworkPolicies, namespace-bound AES-256-GCM encrypted secrets |
| Delivery | Signed OCI images. Source code never shared. Updates delivered as new image versions. |
Detailed infrastructure sizing is provided during the pilot scoping based on the customer's user count, throughput requirements, and retention policy.
09 · Compliance framework alignment
NativeAI Guard creates auditable evidence that AI usage complies with applicable regulations. The following frameworks are addressed:
DSG Art. 6, 12, 16-17Data minimization, records of processing, cross-border transfer control (Swiss Federal Act on Data Protection)Anonymization + audit log
GDPR Art. 25, 32Data protection by design, security of processingPolicy engine + encryption
Art. 47 BankGSwiss banking secrecy. Criminal liability for unauthorized disclosure of client-identifying data.DLP at prompt layer
Art. 321 CP / StGBProfessional secrecy (medical, legal, clergy). Criminal liability for patient/client data disclosure.DLP at prompt layer
Art. 35 VAGInsurance secrecy. Unauthorized disclosure of policyholder data.DLP at prompt layer
Art. 84 KVGHealth insurance data protection. Processing of insured persons' data.DLP + transient processing
FINMA 08/2024AI governance, risk classification, AI inventory and usage monitoring. Supervisory reviews ongoing.Shadow-AI dashboard + audit
FINMA Circ. 2023/1Operational Risks and Resilience (Banks). Requires protection of Client-Identifying Data (CID).DLP at prompt layer
FINMA Circ. 2018/3Outsourcing (Banks and Insurers). External LLM providers require due diligence and Cloud Act risk assessment.Audit trail + monitoring
FINMA SC 05/2020Cyber Incident Reporting. AI-related data leaks qualify as reportable cyber incidents.SIEM + incident export
EU AI Act Art. 9, 14, 15Risk management, human oversight, robustness and cybersecurity for high-risk AI systems.Policy engine + audit trail
DORA Art. 7, 9, 28ICT incident prevention, third-party risk management. LLM providers are ICT third parties.SIEM streaming + audit
NIS2 Art. 21, 23Supply-chain security and incident reporting for AI-related attack vectors.SIEM + incident export
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 6
NativeAI GuardIntegration & Operations
10 · SIEM, SOC and SOAR integration
NativeAI Guard streams structured security events to any SIEM, SOC, SOAR, XDR, or incident management platform. The relevant connector is built and delivered as part of the pilot phase.
Integration protocols
| REST API | Production-ready. Structured JSON audit events for all policy decisions. |
|---|---|
| WebSocket | Production-ready. Real-time event streaming. |
| Syslog (RFC 5424) | Roadmap. UDP/TCP, TLS-wrapped. Developed based on client demand. |
| CEF | Roadmap. Standard format accepted by Splunk, Sentinel, QRadar, ArcSight, Elastic, and others. Developed based on client demand. |
| Webhook | Production-ready. For SOAR/ticketing integration (Jira, ServiceNow, PagerDuty). |
Compatible SIEM and security platforms
SplunkMicrosoft SentinelIBM QRadarGoogle ChronicleElastic SecuritySumo LogicCrowdStrike FalconPalo Alto Cortex XSOARSentinelOneFortinet FortiSIEMRapid7 InsightIDRLogRhythmWazuhGraylogDatadog SecurityJiraServiceNowPagerDutyOpsgenie
11 · Performance
| Policy Engine latency | Low hundreds of milliseconds per request. GPU-accelerated, transient GPU memory processing. |
|---|---|
| Browser Extension | No measurable impact on browser performance in normal operation. |
| Availability target | Custom SLA for Enterprise tier. |
| Auto-scaling | Automatic. Policy Engine scales horizontally based on request volume. |
12 · Operations
| Updates | Continuous, zero-downtime deployments (ArgoCD) |
|---|---|
| Service levels | Response times, patching windows, and availability commitments are agreed per customer and set out in the service agreement. |
| Security Advisors | Marc-Etienne Cortesi (former CISO Baloise, Director Swiss Financial Sector Cyber Security Centre), Alexander Bösch (Deputy CSO SIX Group) |
| Pilot support | Dedicated Success Manager + weekly status calls |
| Priority partner access | Direct access to the engineering team |
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 7
NativeAI GuardPricing
13 · NativeAI Guard Employee DLP: per-user annual subscription
Predictable, per-user pricing. No per-request meters, no consumption-based billing, no hidden fees.
| SME (up to 150 users) | CHF 28,800 flat per year. Core DLP detection, custom policy development, advanced threat analytics, AI governance dashboard, email support. |
|---|---|
| Professional (150 to 2,500) | CHF 16 to CHF 4 per user per month (volume-dependent). All SME features plus SIEM integration, dedicated Customer Success Manager. |
| Enterprise (2,500+) | CHF 4.00 per user per month (volume floor). All Professional features plus on-premises deployment option, 24/7 SOC support, custom SLA. |
14 · NativeAI Guard Agent DLP & Firewall: metered pricing
NativeAI Guard Agent DLP & Firewall is priced on a metered, usage-based model. The volume of traffic between AI agents varies significantly by use case, making per-user pricing impractical. Pricing is scoped individually based on the customer's architecture and expected throughput.
| Pricing model | Metered, usage-based. Billed according to the volume of API traffic inspected by the proxy. |
|---|---|
| Quote | Custom quote on request. Contact NativeAI for a tailored pricing proposal based on your agent architecture and expected traffic volume. |
| Pilot phase | Included in the standard pilot engagement. Traffic profiling during the pilot period informs the production pricing quote. |
15 · Pilot engagement
| Fully managed pilot phase | From CHF 5,000. NativeAI runs everything: implementation, policy configuration, deployment, monitoring, reporting. |
|---|---|
| Standard scope | Up to 30 users for one month. The base price of CHF 5,000 covers this scope. |
| Extensions | Expanding the pilot phase to more users (e.g. 100 or 300) or extending the duration beyond one month increases the pilot fee. Contact us with your requirements to receive a quote for the pilot phase. |
| Duration | Standard: one month passive observation. Optional second month with active blocking and anonymization (priced as extension). |
| Deliverable | Report on data leaks found, shadow-AI visibility, AI usage transparency, and cost efficiency analysis. |
| Pilot fee credit | 100% of the pilot fee is credited toward the Year 1 subscription. |
What is included in every engagement
Implementation: fully included in the pilot fee. Moving to production adds no implementation costs.
Standard base policies: NativeAI provides and manages base compliance policies covering DSG, GDPR, banking secrecy, insurance secrecy, medical secrecy, and other regulatory frameworks. These are tested, optimized, and updated by NativeAI as regulations change. Included in the license, no additional cost.
Custom policy configuration: during onboarding, NativeAI works with the client's team to review their specific data protection needs and create custom policies. Included in the pilot fee.
Client self-service: after onboarding, the client can independently fine-tune and optimize custom policies via the admin console.
SIEM/SOAR integration: the relevant connector is built and delivered as part of the pilot phase.
Support: dedicated Success Manager + weekly status calls + direct engineering access during the pilot phase.
Pilot-to-production transition: switching from pilot phase to full subscription incurs no additional transition or rollout fees.
Red-team validation: included. Every pilot phase includes a red-team "fake user" who tests the deployed policies.
Lighthouse partner discount: companies that start in 2026 and agree to serve as reference customers receive a 20% lifetime discount on the subscription.
There are no consulting fees, no integration fees, and no setup charges beyond the license cost.
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 8
NativeAI GuardCompetitive Comparison
16 · Capability comparison
NativeAI Guard occupies the AI DLP / LLM Firewall category. It protects the AI prompt layer that legacy DLP and network security tools were not designed for. The following comparison reflects the state of these product categories as of August 2026.
| Capability | NativeAI Guard | Legacy DLP (Purview, Forcepoint, Symantec, Trellix) | Cloud AI DLP (Nightfall, Cyberhaven, Harmonic) | Platform AI security (Cisco, Palo Alto, Check Point) |
|---|---|---|---|---|
| LLM prompt-layer DLP | Yes | Partial (Edge-only for Purview) | Yes | Partial (bolted on) |
| Agentic AI / XPIA protection | Yes | No | Partial | Partial |
| Swiss data sovereignty | Yes (own servers, CH) | No (US cloud) | No (US cloud) | No (US cloud) |
| US cloud exposure | No | Yes | Yes | Yes |
| Natural-language policy engine | Yes | No (static rules) | Partial | No |
| PII anonymization (not just block) | Yes | No (block/audit) | Partial | No |
| Browser coverage (Chrome + Edge) | Yes | Edge only (Purview) | Yes | Varies |
| Provider-agnostic LLM coverage | Yes | Restricted | Yes | Varies |
| SIEM/SOC integration | Yes | Yes | Partial | Yes |
| Predictable pricing | Per-user, flat | Per-user + PAYG add-ons | Per-user | Platform bundle |
| On-premises option | Yes (Kubernetes) | Yes | No | Varies |
| No vendor lock-in | Yes (independent) | Microsoft ecosystem | Yes | Platform lock-in |
What NativeAI Guard replaces, and what it runs alongside
Some of the categories below do the same job as NativeAI Guard, so a buyer chooses one of them. Others cover different surfaces and are meant to run alongside it. This section states which is which, and where the boundary falls.
Replaces
| AI DLP and LLM firewalls | Cyberhaven, WitnessAI, Nightfall AI, Aurascape, Harmonic Security, nexos.ai, NeuralTrust, Noma Security, Portal26, Strac, Vallum AI, Lasso Security, Lumia Security, Polymer, Credal.ai, ZeroDrift. Same surface and same job as NativeAI Guard. The differences are Swiss jurisdiction, a natural-language policy engine, and anonymization that leaves the prompt usable instead of blocking it. |
|---|---|
| Platform AI security | Cisco AI Defense (Robust Intelligence), Palo Alto (Protect AI), Check Point Infinity (Lakera), SentinelOne (Prompt Security), Cato Networks (Aim Security), F5 (CalypsoAI), CrowdStrike (Pangea), Zscaler (SplxAI), Proofpoint (Acuvity), Tenable (Apex), Snyk (Invariant Labs), Veeam (Securiti.ai). Each of these bought its AI layer. NativeAI Guard replaces that layer only: their firewall, endpoint and SASE products are untouched. A multi-vendor security stack rarely wants its AI controls locked to a single platform. |
As of August 2026, NativeAI is the only Swiss-headquartered, AI-native pure-play in this category. Lakera and Invariant Labs, the two other Swiss entrants, were both acquired by foreign companies in 2025.
Runs alongside (and in some cases integrates directly with)
| Legacy DLP, CASB and SSE | Microsoft Purview, Forcepoint DLP, Symantec DLP (Broadcom), Trellix DLP, Fortinet DLP, Acronis DeviceLock, Netskope. These enforce on structured channels: email, file shares, endpoints, network traffic, and the Microsoft-native surface. NativeAI Guard enforces on free-form prompt content at the browser layer, across Chrome and Edge, provider-agnostic. Purview does reach unmanaged AI tools through In Transit Protection, though only in Edge for Business, on a consumption meter, using static rules. The practical split: keep them for the structured channels, add NativeAI Guard for the prompt layer. |
|---|---|
| Enterprise AI platforms | Langdock and comparable sanctioned in-house assistants. A platform is a destination: it governs what happens inside it. NativeAI Guard sits on the path and inspects content before it leaves the browser, whatever the destination. A platform is structurally blind to the traffic that never enters it, which is where shadow AI lives. |
| AI red-teaming and model security | HiddenLayer, Cranium AI, Adversa AI, Mindgard. These test and harden models before deployment. NativeAI Guard enforces policy at runtime, on live traffic. Different phase of the lifecycle, no overlap in function. |
NativeAI Guard · Technical Reference · NAI-DOC-001 · v1.2 · 2026-09-21Page 9
NativeAI GuardFAQ
17 · Frequently asked questions
What AI tools does NativeAI Guard support?
Any web-based AI tool is automatically covered. NativeAI Guard operates at the browser layer, so no provider-specific API integrations are needed. This includes ChatGPT, Microsoft Copilot, Claude, Google Gemini, Grok, DeepSeek, Perplexity, DeepL, and over 200 other surfaces. Chrome and Microsoft Edge are supported in production; Firefox and Safari are on the roadmap.
How does NativeAI Guard differ from Microsoft Purview DLP?
They overlap in one narrow place and are complementary everywhere else, so most customers run both. Purview protects structured channels (Exchange, SharePoint, Teams, Copilot) within the M365 ecosystem, with GenAI blocking limited to Edge for Business and static-rule policies. NativeAI Guard covers the LLM prompt layer across Chrome and Edge, uses a natural-language policy engine with semantic understanding, provides utility-preserving anonymization (not just blocking), and runs on NativeAI's own servers in Switzerland with no US CLOUD Act exposure.
Does NativeAI Guard require API access to the LLM provider?
No API access is needed. For employee AI protection, NativeAI Guard works as a browser extension that intercepts content before it reaches any AI tool. For agentic AI protection (internal agents, RAG pipelines, chatbots), it deploys as a drop-in API proxy between your systems and the LLM APIs.
How is data residency handled?
All processing stays in Switzerland, on NativeAI's own servers in Lausanne. No public cloud (no AWS, Azure, or GCP). Data that never reaches a US provider cannot be compelled under the US CLOUD Act or FISA 702. For on-premises deployments, all processing stays within the customer's own infrastructure.
What are the on-premises deployment requirements?
Kubernetes with NVIDIA H100 or H200 GPU nodes is required. Multiple GPU nodes are needed depending on user count and throughput. This GPU infrastructure is the primary cost driver for on-premises deployments: customers without existing GPU infrastructure should budget accordingly. Also required: persistent storage for audit logs and network connectivity. Source code is never shared.
Is there an implementation fee beyond the license cost?
No. There are zero implementation fees. Implementation, policy configuration, deployment, and onboarding are fully included in the pilot fee. Moving from pilot phase to production adds no costs. 100% of the pilot fee is credited toward the Year 1 subscription.
How does NativeAI Guard detect prompt injection attacks?
GPU-accelerated detection models inspect every prompt and response in real time. For prompt injection (XPIA), it detects hidden adversarial instructions embedded in external data. For jailbreaks, it identifies attempts to bypass agent guardrails. Detection models are updated continuously.
Can NativeAI Guard integrate with our existing SIEM?
Yes, via REST API and WebSocket real-time streaming. Structured JSON audit events for all policy decisions. Syslog and CEF export on the roadmap. Compatible with Splunk, Microsoft Sentinel, IBM QRadar, Google Chronicle, Elastic Security, CrowdStrike Falcon, Palo Alto Cortex, and many others.
What is the performance impact on end users?
Imperceptible. The GPU-accelerated Policy Engine processes content in the low hundreds of milliseconds per request. Since AI requests already carry multi-second response times, NativeAI Guard's overhead is not noticeable. The Browser Extension has no measurable impact on browser performance.
What compliance frameworks does NativeAI Guard support?
DSG, GDPR, EU AI Act, four FINMA instruments, DORA, NIS2, and Swiss sector secrecy statutes. FINMA coverage: Guidance 08/2024 (AI governance), Circular 2023/1 (CID protection), Circular 2018/3 (LLM outsourcing), SC 05/2020 (cyber incident reporting). Secrecy statutes: Art. 47 BankG, Art. 321 CP, Art. 35 VAG, Art. 84 KVG. Aligned with ISO 27001; certification in progress.
NativeAI Guard
Next step
Book a demo of NativeAI Guard. 45 minutes, live software. See what is possible today, what a safety net for AI use looks like, and how easy it is to protect the AI use of your employees and your AI agents. Book at nativeai.ch or write to [email protected].
NativeAI Sàrl
EPFL Innovation Park, Bâtiment C
1015 Lausanne, Switzerland
nativeai.ch · linkedin.com/company/nativeai-ch
1015 Lausanne, Switzerland
nativeai.ch · linkedin.com/company/nativeai-ch
Certifications & memberships
Aligned with ISO 27001; certification in progress
Swiss AI member · La Forge, EPFL Innovation Park
Swiss AI member · La Forge, EPFL Innovation Park
© 2026 NativeAI · Made in Switzerland · All data processed on our own servers in Lausanne, under Swiss law · NAI-DOC-001 · v1.2 · 2026-09-21