← All posts

Compliance & Regulation

463 posts on compliance & regulation.

BSI AIC4 Criteria: What Runtime AI Traffic Can Prove

The BSI AIC4 criteria catalogue is a full-lifecycle assurance framework for cloud AI services and an extension to BSI C5. Runtime model traffic can support request logging, request monitoring, and incident reconstruction. It cannot prove provider training, validation, data quality, explainability, bias controls, or operating effectiveness across the service.

ai-compliancecloud-securityauditai-governanceforensic-auditpolicy-enforcement
Read post →

EDPS Generative AI Orientations: Evidence for EU Institutions

The EDPS generative AI orientations apply Regulation (EU) 2018/1725 to EU institutions, bodies, offices, and agencies. They call for documented roles, DPO involvement, risk assessment, data minimisation, traceability, security, and regular review. Runtime records can support that work, but they cannot decide legal basis or prove how a provider trained its model.

ai-governanceai-complianceregulationgdprauditpolicy-enforcement
Read post →

NIST AI 100-4 Makes Synthetic Content a Layered Control Problem

NIST AI 100-4 separates digital content transparency into provenance tracking and synthetic content detection, then documents the limits of watermarking, metadata, automated classifiers and human review. The report supports a layered control plan with context-specific evaluation, explicit uncertainty and clear ownership. This guide maps the NIST approaches to operating evidence and draws the boundary around routed LLM HTTP traffic.

nistnist-ai-rmfai-securityai-governanceauditpolicy-enforcement
Read post →

AIUC-1 Turns AI Agent Security Into Auditable Requirements

AIUC-1 defines requirements for AI agent security, safety and reliability, with accredited-auditor validation and recurring evaluations. Its security section covers adversarial testing, endpoint protection, agent permissions, user access, deployment environments, information leakage and generated code. This guide turns those requirements into an evidence plan and marks the limited set of controls an authenticated HTTP policy point can support.

ai-securityai-governanceai-complianceagentic-aiauditpolicy-enforcement
Read post →

Standard Contractual Clauses for AI Vendors: Select the Contract by Route

The European Commission modernized its standard contractual clauses on 4 June 2021. AI procurement teams still need to match the contract to the parties and roles, plus the data route and destination involved in each vendor service. This guide organizes SCC selection around an evidence pack for LLM vendors, including access locations, subprocessors, data classes, approved endpoints, supplementary measures, suspension, change control, and retention.

gdprai-complianceregulationai-governancecloud-securitycompliance
Read post →

Schrems II for LLM Transfers: Build the Assessment Around the Route

Schrems II preserved standard contractual clauses while requiring protection that remains essentially equivalent after personal data leaves the EEA. For LLM deployments, the work starts with the actual request route, including cloud processing and remote access outside the EEA. This guide applies the EDPB six-step method to model traffic and separates legal assessment from the technical evidence an authenticated HTTP route can produce.

gdprai-complianceregulationai-governancecloud-securityaudit
Read post →

Kimi Audit Logs: Separate Control-Plane Events From Model Requests

Kimi now exposes organization audit events and per-request billing details through separate Admin API feeds. The first records changes to credentials, members, projects, login events, and settings. The second identifies the model, project, API key, token counts, and a client-supplied request ID. A defensible review must correlate both feeds with an independent record of the person or agent, prompt classification, and policy decision on the routed model request.

ai-governanceauditllm-securityidentity-and-authorizationpolicy-enforcement
Read post →

AI SLA Template: Route-Level SLOs and Claim Evidence

An AI service-level agreement becomes operational only after its provider definitions are translated into route-level objectives and claim evidence, with failover rules defined for each route. This guide turns AWS Bedrock and Google Vertex AI terms into a worksheet for authenticated LLM traffic without confusing provider credits with application reliability.

ai-governanceai-complianceai-securityauditarchitecturecloud-security
Read post →

Preventing Woke AI Executive Order: A Federal Procurement Evidence Map

Executive Order 14319 and OMB Memorandum M-26-04 turn two Unbiased AI Principles into federal LLM procurement requirements. Agencies need vendor disclosures, contract terms, reporting procedures, and evidence that procured models meet the stated principles. Procurement records and observable output evaluations serve different purposes, so a defensible program preserves both.

ai-governanceai-procurementai-compliancefederal-ai-policyllm-evaluationpolicy-enforcement
Read post →

EU AI Act Article 4 AI Literacy Needs Role-Specific Evidence

EU AI Act Article 4 requires AI providers and deployers to support AI literacy among staff and other people who operate or use AI systems on their behalf. The Commission points organizations toward role, knowledge, experience, training, system context, and affected people. A defensible program maps real AI systems to real roles, teaches system-specific risks, checks applied competence, and preserves an internal record of the work.

eu-ai-actai-governanceai-compliancecomplianceregulation
Read post →

AI Indemnification Depends on the Controls You Can Prove

AI indemnification clauses can cover defined intellectual-property claims while excluding outputs, services, or conduct outside stated conditions. Microsoft ties its Customer Copyright Commitment to required mitigations. Google limits its generated-output protection to listed services and excludes several customer-controlled scenarios. A claim-ready program preserves the contract version, covered route, required control state, output lineage, and notice record for each disputed output.

ai-governanceai-complianceauditpolicy-enforcementforensic-audit
Read post →

ISO/IEC 22989 Terminology Needs an Operational Dictionary

ISO/IEC 22989:2022 establishes concepts and terminology for artificial intelligence. Organizations still need an operational dictionary that maps those terms to their own systems, use cases, actors, components and evidence. This guide turns terminology into controlled identifiers and record fields, shows how to handle competing legal definitions, and limits runtime gateway evidence to authenticated HTTP traffic deliberately routed to LLM endpoints.

ai-governanceai-compliancecomplianceauditiso-42001
Read post →