Topic
Security and governance
Credentials, data paths, model access policy, audit and the evidence a security or procurement review actually asks for.
13 articles · Last updated
Articles
Security & governance
AI gateway keys: environments, limits, rotation, off switch
The key your application holds should never be a provider key. Here is how to scope gateway keys by environment, provider and network, cap what each can spend, rotate without downtime, switch one off, and respond when one leaks.
Guide12 min read
Security & governance
LLM guardrails at the gateway: checks before and after
A guardrail is a check with a decision attached, not a promise of safety. Here is which checks belong in the application, which at a gateway and which at the provider, what each costs, and how to decide what happens when one cannot run.
Guide13 min read
Security & governance
Jev for LLM guardrails and model routing
Every production LLM call needs decisions around it: is this prompt safe, which model should answer, is the reply acceptable. Those checks run on every call, so they have to be fast and small next to the call itself. This is how to build them with TypeSafe's Jev, and what they cannot do.
Guide7 min read
Security & governance
What security reviewers ask about AI gateways
An AI gateway sits in the path of every model call, so it gets a harder review than most SaaS. These are the questions reviewers ask, what separates a good answer from a weak one, and how to tell a mechanism from a promise.
ChecklistUpdated 7 min read
Security & governance
BYOK gateway architecture: sealing, scoping and rotating keys
A BYOK gateway holds credentials that spend money on accounts it does not own. This guide covers the four credential types, authenticated encryption bound to a row, write-only secrets, fail-closed scoping and three different kinds of rotation.
Guide9 min read
Security & governance
A threat model for an LLM gateway, mapped to OWASP
An LLM gateway is a new asset in the architecture, not just a proxy. This analysis names what it holds, where its trust boundaries are, and maps the OWASP 2025 list to what a gateway can actually do about each risk — including the risks it cannot touch.
Analysis8 min read
Security & governance
A prompt and payload logging policy for LLM traffic
Every team running LLM traffic ends up with a logging policy, written or not. This checklist makes it explicit: what is recorded always, what is captured only on purpose, who can read it, how long it lives and what leaves for other systems.
Checklist7 min read
Security & governance
Prompt injection in production: defense in depth
Prompt injection is not a bug one filter can fix. This guide lays out the layers that make it survivable in production — with code for authorizing tool calls outside the model, and an honest account of what a gateway can and cannot contribute.
Guide8 min read
Security & governance
Least privilege for provider admin and usage keys
Cost and usage reporting needs an admin credential, and at most providers that credential can do far more than read a report. This guide compares what OpenAI, Anthropic and AWS Bedrock let you scope, and how to hold the key you cannot narrow.
Guide7 min read
Security & governance
Enterprise AI gateway checklist: security, control, and scale
An enterprise gateway is a control boundary, not a model menu. Use this checklist to evaluate access, data handling, reliability, auditability and economics.
ChecklistUpdated 2 min read
Security & governance
How to secure LLM API keys in production
A key is not just a secret; it is spend and data-path authority. Scope it, store it correctly, monitor it, cap it, rotate it, and rehearse revocation.
BriefUpdated 1 min read
Security & governance
Model access governance for fast-moving AI teams
Governance should make the safe path easy: approved models by workload, explicit exceptions, traceable changes, and fast evidence-based evaluation.
BriefUpdated 1 min read
Security & governance
LLM gateway security architecture: one enforceable boundary
A gateway adds a party to the data path, but it also creates one place to enforce credentials, model policy, budgets, logging, and provider access.
BriefUpdated 1 min read