Most recent
navigate open esc close Corpus index built 2026-06-07 23:58 UTC

§ THE STACK / GATEWAY LAYER

LLM Gateways

LiteLLM, OneAPI, model routing

Routes the request, attaches retrieved context, mediates between user and model.

What it is

An LLM gateway is a reverse proxy for model APIs. The operator wires up keys for OpenAI, Anthropic, Google, Mistral, their own Ollama box, and a handful of fine-tunes; the gateway exposes a single OpenAI-compatible endpoint and handles routing, rate-limiting, fallback, observability, and cost accounting. LiteLLM is the Python-native one (most common in research); OneAPI is the Go/Chinese-ecosystem one (most common in commercial deployments). Portkey, Helicone-Proxy, and APISIX-AI sit in the same niche.

What goes wrong

The gateway holds the operator’s entire AI billing relationship. If it’s exposed without auth, an attacker can route arbitrary prompts through any of the configured providers: burning the operator’s quota, exfiltrating embedded prompts that may contain customer data, and racking up usage charges on premium models. Worse: the admin panel typically lists every model alias, the keys behind them, and the per-user/per-team budget. The attacker learns the operator’s whole AI org chart before issuing a single request.

How we test

We confirm the gateway by its /v1/models response shape (LiteLLM’s is distinct from a vanilla OpenAI proxy), then check /health/readiness and /key/info for admin-key reachability. The key endpoint, when unauthenticated, returns the operator’s full virtual-key inventory including budget caps and team assignments. We do not issue paid completions. The catalogue is enough to demonstrate the quota-drain risk and identify the operator.

Receipts

Research

Every survey, case study, and disclosure we've published that touches this layer of the stack. Counts on the cells above tally these directly.

Cross-cloud surveys

8
Survey Jun 6, 2026

Cat-05: LiteLLM Gateway Survey — Open Proxies Exposing Commercial LLM API Keys

The hunt started with a single Shodan dork: http.title:"LiteLLM" port:4000. It returned 2,219 results in under a second.

Read →
Survey Jun 1, 2026

AI Gateways Population Survey: Cat-32 (2026-06-01)

An AI gateway sits in front of every upstream LLM provider an operator uses. It holds the OpenAI key, the Anthropic key, the Gemini key, the DeepSeek key. All in one process. That is the point of the…

Read →
Survey May 19, 2026

LLM Safety / Guardrail / Policy Engine population survey

The auth-on-default thesis predicts that products which ship without authentication will appear at population scale with the unauth posture intact. The LLM safety / guardrail / policy layer is the inv…

Read →
Survey May 17, 2026

LLM gateway / proxy population survey, 2026-05-17

We surveyed the public-facing LLM gateway / API-proxy population: LiteLLM, Helicone, Portkey, OneAPI, NewAPI, OpenRouter self-host. A LLM gateway sits between an application and one or more upstream L…

Read →
Survey May 11, 2026

VisorBishop iter-5: LiteLLM Proxy + Argilla + Promptfoo (gateway + annotation + eval tiers)

NuClide Research · 2026-05-11

Read →
Survey May 11, 2026

VisorBishop iter-6: Full LiteLLM 5,391-host population sweep (283 unauth LLMjacking primitives)

NuClide Research · 2026-05-11

Read →
Survey May 11, 2026

VisorBishop Phase 5: Three primitives that turn 492 critical hosts into an impact narrative

NuClide Research · 2026-05-11

Read →
Survey May 1, 2026

LLM Gateways / OpenAI-Compatible Proxies: Cross-Cloud Survey (2026-05)

LLM gateway / OpenAI-compat proxy products sit between LLM applications and upstream providers. They normalize multiple provider APIs (Anthropic, OpenAI, Cohere, Together, etc.) behind a single OpenAI…

Read →