Process · Security & Compliance

Secure, Governed, & Compliant by Design

Frontier Foundry systems operate inside regulated production environments with structured controls across infrastructure, execution, and oversight.

Controls apply at four layers: infrastructure isolation, policy enforcement, deterministic execution and monitoring. Every operation is logged from input to output. Customers retain ownership of the models and weights trained in their environment.

Compliance alignment

Frontier Foundry has completed a SOC 2 Type II examination covering a 2026 examination period. The report and auditor details are available under NDA.

Deployments align with the regulatory frameworks applicable to each institution.

Attestation: SOC 2 Type II report (under NDA).

Frameworks our deployments are designed to support: GDPR, CCPA, BSA/AML and KYC, HIPAA, NIST Cybersecurity Framework, NYDFS Part 500.

Assurance

What your reviewers will ask for.

  • SOC 2 Type II report (under NDA)

    Frontier Foundry has completed a SOC 2 Type II examination covering a 2026 examination period. The report and auditor details are available under NDA.

  • Model risk management (SR 26-2)

    Built for model risk programs under the agencies' revised model risk guidance (SR 26-2, April 2026, which replaced SR 11-7 and SR 21-8). Kundi's deterministic decision path, versioned models and a full audit trail provide the evidence validators need for conceptual soundness, ongoing monitoring, and outcomes analysis. A validation pack is available on request.

    See what the validation pack includes →

  • NIST AI Risk Management Framework

    Our deployments map to the NIST AI RMF functions — Govern, Map, Measure, and Manage — with a control-by-control crosswalk available on request.

  • EU AI Act

    Designed to support high-risk-system obligations under the EU AI Act — record-keeping, transparency, and human oversight — ahead of the December 2, 2027 application date for Annex III systems.

  • Federal deployments

    Our platforms run on-premises or fully air-gapped inside an agency's existing authorization boundary, so they inherit its security controls. Qualified government buyers can request details under appropriate agreements.

  • Post-quantum cryptography

    Post-quantum cryptography protects data, keys, and connections across all of our platforms, using the FIPS-approved algorithms ML-KEM, ML-DSA, and SLH-DSA (FIPS 203, 204, and 205), each verified byte-for-byte against official test vectors.

    See how we verify it →

The security stack

Security is implemented as a layered system.

Infrastructure

Deployment Boundary Enforcement

  1. 1

    Data remains inside customer-controlled environments

  2. 2

    AI runs within defined network boundaries

  3. 3

    Sensitive information does not leave approved systems

Technical details:

  • Deployment within customer-managed cloud (AWS, Azure, GCP), on-prem, or fully air-gapped environments
  • AES-256 encryption at rest
  • TLS 1.2+ encryption in transit
  • Zero Trust access architecture with enforced multi-factor authentication (MFA)
  • Virtual network segmentation and isolated execution subnets
  • Runtime isolation by component: applications can run in containers (Podman, Compose, Helm for Kubernetes) or as native packages; Tarn installs natively with no containers.
  • Data residency configuration based on jurisdiction
  • Customer retains ownership of trained models, weights, and improvements derived from their data
Governance

Policy & Access Enforcement

  1. 1

    Access and execution governed at system level

  2. 2

    Policies are enforced consistently across models and workflows

  3. 3

    System activity remains visible and controlled

Technical details:

  • Integration with enterprise IAM providers (Azure AD, Okta, SAML, SSO)
  • Granular RBAC across datasets, models, and workflows
  • Policy enforcement at execution layer
  • Version-controlled configuration management
  • Administrative logging and activity monitoring
Execution

Deterministic Runtime Control

  1. 1

    Model execution is deterministic and traceable

  2. 2

    Inputs and outputs are logged

  3. 3

    Systems operate continuously with monitoring in place

Technical details:

  • Reproducible execution pipelines with version-locked artifacts
  • Structured input/output trace logging
  • Model drift detection with threshold alerts
  • Performance monitoring dashboards (latency, accuracy, throughput)
  • Automated restart and health checks
Audit & Oversight

Traceability & Reviewability

  1. 1

    Outputs are review-ready

  2. 2

    Decisions are inspectable

  3. 3

    Execution history is preserved

Technical details:

  • Exportable audit logs (JSON/CSV)
  • Time-stamped execution history tied to user and model versions
  • Structured reasoning artifacts attached to outputs
  • Evidence retention configuration aligned with regulatory timelines
  • Reporting templates for internal audit and regulator-facing review

Regulatory alignment by market

Security posture adapts to industry-specific requirements.

  • Banking

    Supports BSA / AML, KYC, call report obligations, and financial data governance standards

  • Healthcare

    Supports HIPAA and PHI governance requirements within secure environments.

  • Government

    Deployable under agency-specific regulations and jurisdictional data policies.

  • Asset Management

    Supports SEC, FINRA, NYDFS Part 500, BSA / AML for advisers, and fund administration governance.

Documentation & due diligence

Security documentation is available during evaluation.

  • Architecture diagrams
  • Deployment topology maps
  • Governance control matrices
  • Data flow documentation
  • Model monitoring procedures
  • Compliance mapping summaries
Quantum-Resilience

Quantum-Resilient Out of the Box

Every Frontier Foundry system uses post-quantum cryptography today: ML-KEM for key establishment and ML-DSA and SLH-DSA for signatures (FIPS 203, 204 and 205), verified against NIST test vectors. Algorithms are abstracted so they can be rotated as standards change.

  • PQC-Native Key Establishment and Signatures

    Post-quantum key establishment and digital signatures are part of the platform's cryptographic layer, extending the encryption already enforced in transit and at rest.

  • Crypto-Agile Architecture

    Cryptographic primitives are abstracted so algorithms can be rotated as NIST standards evolve, without re-architecting the systems that depend on them.

  • Built Against Harvest-Now, Decrypt-Later

    Long-shelf-life regulated data is treated as already exposed to future decryption, so it is protected under post-quantum cryptography today rather than after a quantum-capable adversary arrives.

Why It Matters

Public estimates of the qubits needed to break RSA-2048 fell about 20-fold from 2019 to 2025 (Gidney and Ekerå, 2019; Gidney, 2025). The June 22, 2026 executive order sets federal deadlines of December 31, 2030 for post-quantum key establishment and December 31, 2031 for post-quantum signatures on high-value assets and high-impact systems. Under a harvest-now, decrypt-later threat model, long-shelf-life data captured today should be treated as already exposed. Frontier Foundry systems use post-quantum algorithms (ML-KEM, ML-DSA, SLH-DSA) today, so customers have no PQC migration to schedule for our platform. Our founder called the threat window in his article “Has the Battle for Quantum Supremacy Already Been Lost?”

Read the founder's analysis on Substack
Related