Security & Trust
Security is architecture,
not a feature.
Ulfberht is built for environments where a data breach, a compliance gap, or a single unauthorized transmission carries regulatory and operational consequence. Every design decision reflects that.
Principles
Four non-negotiables
These are not product differentiators. They are the conditions under which we agreed to build this platform.
Principle 01
Zero Trust by Default
Every agent-to-agent communication is verified. No implicit trust between components, sessions, or services. Trust is granted by explicit cryptographic scope, not by position in the pipeline.
Every message treated as potentially compromised until verified. No lateral movement possible by design.
Principle 02
Air-Gap Ready
Full on-premises deployment with zero external calls. Your data, your models, your network boundary. Ulfberht operates entirely within your infrastructure when deployed on-premises -- nothing leaves your environment.
Designed for classified, clinical, and highly regulated environments where external connectivity is not permitted.
Principle 03
Cryptographic Integrity
Hash-linked audit trails. Tamper-evident governance records. Every verification decision is cryptographically sealed. Zero-knowledge compliance proofs allow you to demonstrate regulatory adherence without exposing the underlying data.
Auditors receive proofs of process -- not raw data. Patient records, proprietary inputs, and confidential business data stay private.
Principle 04
No Training on Your Data
Your AI outputs are verified, not stored. We do not train models on customer data. We do not retain outputs beyond the active verification session. Your data is not Ulfberht's product. It never was. It never will be.
This applies to all deployment modes: cloud, on-premises, and air-gapped. No exceptions by tier, contract size, or use case.
Certification Status
Built for regulated environments
Ulfberht is architected for the frameworks your compliance and legal teams already operate under. Certifications are pending. We note current design intent and architecture posture -- not certifications we have not yet obtained.
SOC 2 Type I
SOC 2 Type I certification pending. Security, availability, and confidentiality trust service criteria addressed in architecture.
HIPAA
HIPAA compliance pending. PHI handling, minimum necessary access, and audit logging designed to HIPAA Security Rule requirements.
EU AI Act
EU AI Act compliance pending. Architecture addresses Article 14 (human oversight) and Article 15 (accuracy and robustness). Oversight tier classification and claim verification provide the technical mechanism these articles require.
NIST AI RMF
NIST AI RMF alignment pending. Verification layers address the four RMF functions: GOVERN, MAP, MEASURE, and MANAGE. Behavioral pattern detection and audit trails address measurable risk tracking requirements.
FedRAMP
FedRAMP authorization pending. Infrastructure architecture addresses FedRAMP control families. On-premises and air-gapped deployment options support federal environment requirements where cloud authorization is restricted.
ISO 27001
ISO 27001 certification pending. Information security management controls address Annex A requirements. Access control, cryptography, operations security, and supplier relationships addressed in system design.
GDPR / CCPA
GDPR and CCPA compliance pending. Data subject rights architecture, PII detection in AI outputs, right to erasure support, and data residency options designed to support these obligations.
ISO 42001
ISO 42001 certification pending. AI management system requirements addressed. Risk assessment, impact analysis, and continual improvement processes designed to align with the ISO 42001 AI governance framework.
"Pending" indicates architecture intent and active pursuit, not formal certification status. Formal certification documentation and security questionnaire responses available upon request.
Data Handling
Where data flows -- and where it stops
Ulfberht processes AI outputs through a verification pipeline. Data flows in, governance decisions flow out. Nothing is stored beyond the active session.
AI output received
Claims extracted, context read, session scoped
Six-layer verification pipeline
Data processed in-memory within your environment
Verified result delivered
Confidence score, claim status, audit record written
Session closes. Nothing retained.
No persistent storage of AI input or output data
Encryption
In transit and at rest
In transit: TLS 1.3 minimum. Older protocol versions disabled.
At rest: AES-256 encryption. Keys managed per customer in cloud deployments.
Customer-managed encryption keys available for enterprise tier.
Data Residency
Your data, your region
Cloud deployment: region selection available (US, EU, AP).
On-premises: data never leaves your physical infrastructure.
No cross-region data transfer without explicit customer authorization.
Individual Rights
Right to erasure
Data subject access request (DSAR) process supported.
Deletion requests honored across all stored configuration and audit log data.
PII detection in AI outputs flags data that may require handling under GDPR / CCPA.
Infrastructure
Three deployment modes. One security posture.
Every deployment mode maintains the same security architecture. The difference is boundary ownership, not security level.
Option A
Cloud
Hosted on SOC 2 pending infrastructure. Managed by Ulfberht. Customer configures access controls, selects data region, and manages encryption keys.
Option B
On-Premises
Deployed within your infrastructure. Your network, your servers, your policies. Ulfberht provides the software layer. Your team controls the environment.
Option C
Air-Gapped
Fully isolated deployment. Zero external calls. Designed for classified, clinical, and critical infrastructure environments where network isolation is a hard requirement.
Access Control
Identity and authorization
Role-based access control (RBAC) with least-privilege defaults. Granular permission scopes per user, team, and integration.
SSO / SAML 2.0 support for enterprise identity providers. SCIM provisioning for automated user lifecycle management.
MFA enforced for all administrative and privileged access paths. Hardware key support available.
Network Security
Isolation by default
VPC isolation per customer deployment. No shared networking between tenants in cloud mode.
No public-facing administrative endpoints. Management interfaces restricted to private network access.
Egress filtering enforced. Outbound connections explicitly allowlisted -- no implicit internet access from the verification pipeline.
Monitoring
Real-time visibility
Structured security event logging with immutable audit trail. Every access, every action, every governance decision logged.
Anomaly detection on authentication patterns and API usage. Threshold alerts for unusual access volumes or timing.
Customer-accessible security event export for SIEM integration. Splunk, Datadog, and custom webhook delivery supported.
Responsible AI
The governance system has governance.
A verification platform that cannot be verified is not a security product. Ulfberht's governance layer applies the same scrutiny to its own evaluators that it applies to your AI systems.
This is not a philosophical position. It is an architectural requirement. If our evaluators drift, your AI outputs receive false confidence scores. That failure mode is designed out, not monitored for.
Canary Testing
Known-failure inputs are continuously passed through the governance pipeline. When the system begins approving outputs it should flag -- a drift indicator -- automatic recalibration triggers before any customer output is affected.
Explainable Governance Decisions
Every governance decision includes a reasoning chain. Why was this claim flagged? Which behavioral pattern was detected? What evidence supports the confidence score? No black-box verdicts. Your compliance team can read every decision.
No Governance Bypass
No agent within the platform can modify its own governance constraints. Permission to skip verification cannot be granted at runtime, only at the deployment configuration level -- and only to roles with explicit administrative authority. Bypassing governance is an auditable event, not a silent override.
Security Documentation
Request our security package.
For enterprise procurement and compliance review: security questionnaire responses, architecture documentation, data flow diagrams, and subprocessor list available upon request under NDA.
Formal certification documentation and security questionnaire responses available upon request. "Pending" reflects certification intent and current architecture posture at time of writing and does not constitute a certification claim or compliance guarantee. Contact your legal and compliance team to assess suitability for your regulatory environment.