← All posts

Platform & Architecture

338 posts on platform & architecture.

GitHub Copilot Audit Logs: What They Record

GitHub Copilot audit logs record enterprise and repository activity. They do not preserve the code context, originating identity, policy decision, and model route for each AI request. This article shows the evidence record a Copilot review needs.

auditforensic-auditllm-securityai-securitydevsecopsidentity-and-authorization
Read post →

OpenAI Agent Builder Security: Injection Controls

OpenAI Agent Builder security controls for prompt injection, untrusted data, structured outputs, MCP tool approvals, and request authorization.

ai-securityagentic-aiprompt-injectionidentity-and-authorizationinline-enforcementllm-security
Read post →

AWS Bedrock Audit Logs Explained

How Amazon Bedrock model invocation logging and CloudTrail record AI calls, and why the identity.arn field is not a natural-person record.

ai-securityauditforensic-auditllm-securitycompliancearchitecture
Read post →

Microsoft Sentinel AI Logs Need a Deliberate Data Contract

Microsoft Sentinel AI logs work when the team defines source ownership, a stable event contract, table placement, correlation IDs, analytics rules, and retention before ingestion. Azure OpenAI resource logs provide service telemetry. Independent gateway decisions add authenticated identity and policy evidence. Raw prompts and responses should follow a narrower content path.

ai-securityforensic-auditauditllm-securitycloud-security
Read post →

RBAC for LLMs Needs a Route and Action Matrix

RBAC for LLM systems should decide which identity may use a model route, with which data class, for which action. Cloud roles establish access to provider resources, but an enterprise policy matrix still has to govern prompts, tools, routes, and responses at request time. Denial tests and independent decision records make that matrix enforceable.

identity-and-authorizationpolicy-enforcementllm-securityzero-trustarchitecture
Read post →

LLM Cost Management Starts With a Normalized Usage Ledger

LLM cost management needs a normalized request ledger before it needs another dashboard. Provider usage reports expose different identity, model, cache, and service-tier fields. A useful control loop joins those records to application identity, applies versioned prices, enforces budgets before requests run, and reconciles estimates against provider billing data.

llmai-governancearchitecturepolicy-enforcementcloud-security
Read post →

API Key Rotation for LLM Systems: A Runbook That Avoids Downtime

LLM API key rotation is a controlled credential replacement, not a console click. The runbook starts with inventory, creates a restricted replacement, deploys it through a dual-key window, validates every workload, revokes the old key, and preserves a tested rollback path. This guide also separates provider credential controls from HTTP model-request policy and states the stolen-key boundary plainly.

ai-securityidentity-and-authorizationzero-trustpolicy-enforcementllm-securityarchitecture
Read post →

RAG vs Fine-Tuning Security: Two Data Paths, Two Control Plans

RAG supplies retrieved content to a model at request time, while fine-tuning adapts model weights through additional training. That difference changes the security plan for poisoning, authorization, deletion, provenance, evaluation, rollback, and runtime inspection. This comparison maps each control to the corpus, training pipeline, model registry, retrieval service, or HTTP model boundary that can actually enforce it.

ai-securityllm-securityprompt-injectionidentity-and-authorizationpolicy-enforcementarchitecture
Read post →