How does discovery work without installing anything on endpoints?

It reads the egress logs your network already produces. DNS and HTTP logs go into the engine, get matched against a curated set of 16+ AI provider domains, and come out as an inventory of which systems are calling which providers. The trade-off is honest: log-based discovery sees what egresses your network, not what happens inside a laptop that never touches your DNS.

What does dns-redirect mode require?

A client-trusted proxy CA. Redirecting provider domains to the gate at the DNS level means the proxy terminates TLS for those domains, and clients must trust its certificate authority. That’s a real deployment decision with real security implications, and we’d rather you make it knowingly than discover it mid-rollout.

What happens once a shadow system is found?

It becomes an asset: it appears in the topology, gets scored on the 5×5 risk matrix, and shows up in posture alongside registered workloads. From there you choose the path — move it behind the gate with a base-URL change, catch it estate-wide with dns-redirect, or accept and document the risk. Discovery gives you the decision; it doesn’t make it for you.

Shadow-AI discovery

You can’t govern traffic you can’t see.

The AI usage that worries your GRC team most is the usage nobody registered: the copilot in a browser, the script with a personal API key. AnchorGate ingests the DNS and HTTP egress logs you already have and inventories every call to known AI provider domains — then gives you a path to bring that traffic through the gate.

Get started — it's free View on GitHub

The mechanisms

Discover, then govern — in that order

An inventory you can’t act on is trivia. Discovery here feeds the same asset model, risk register, and gate as everything else.

Discovery from logs you already have

No agents to deploy: feed the engine your DNS and HTTP egress logs and it matches them against 16+ AI provider domains, building an inventory of the systems calling AI providers — including the ones nobody registered.

From discovered to governed

dns-redirect mode routes unmanaged clients through the gate at the network level, so estate-wide traffic crosses policy without every team changing code. Honest caveat: interception requires a client-trusted proxy CA — we say that upfront.

Into the risk register, not a report

Discovered systems land in the same asset inventory and risk model as your registered workloads — scored on the 5×5 matrix, visible on the posture topology, one dashboard for both halves of your estate.

What discovery surfaces

The estate you have, not the estate you registered

Egress logs don’t lie: every copilot, notebook, and cron job that calls an AI provider resolves a domain on your network first. Matching those logs against known provider domains turns “we think some teams use ChatGPT” into a list with names on it.

Works from DNS and HTTP logs you already collect — no per-host agents.
Each discovered caller becomes an asset, scored on the same 5×5 risk matrix.
The remediation path is built in: one base-URL change, or dns-redirect estate-wide.

anchorengine · shadow-ai discovery source: egress logs

data-science-notebooks → api.openai.com

Unregistered caller — 1,204 requests this week, personal API key suspected.

unregistered → added to asset inventory

vendor-browser-extension → api.anthropic.com

New provider domain for this host group — first seen 3 days ago.

unregistered risk: scored 5×5

support-copilot → gate :8080

Registered and routed through the gate — every request on the signed ledger.

governed

16+ provider domains matched 2 systems to bring behind the gate

In the box

What ships today

Ingests DNS and HTTP egress logs — no per-host agents
Matches against 16+ AI provider domains
Inventories callers nobody registered — copilots, scripts, notebooks
Discovered systems join the asset topology and risk register
dns-redirect mode brings unmanaged clients through the gate
Requires a client-trusted proxy CA for interception — stated, not hidden
verify mode as the observe-only on-ramp before enforcement

As shipped in AnchorGate v0.1 — every line above is checkable against the repository.

16+: AI provider domains matched in egress-log discovery
3: interception modes: full · verify · dns-redirect
0: per-host agents required for discovery

Provider domain set as shipped in AnchorGate v0.1.

Works with the rest of the gate

Compliance Posture

Control catalogues for the EU AI Act, ISO/IEC 42001, and SOC 2 map directly onto decision evidence. Posture is computed from what the gates actually did — not from a policy document.

Learn more →

Policy Enforcement

A deterministic policy DSL — IF condition THEN redact, alert, or block — compiled, human-approved, and evaluated before a request leaves. No LLM judging LLMs: the same input always gets the same verdict.

Learn more →

FAQ

Shadow-AI discovery questions

How does discovery work without installing anything on endpoints?: It reads the egress logs your network already produces. DNS and HTTP logs go into the engine, get matched against a curated set of 16+ AI provider domains, and come out as an inventory of which systems are calling which providers. The trade-off is honest: log-based discovery sees what egresses your network, not what happens inside a laptop that never touches your DNS.
What does dns-redirect mode require?: A client-trusted proxy CA. Redirecting provider domains to the gate at the DNS level means the proxy terminates TLS for those domains, and clients must trust its certificate authority. That’s a real deployment decision with real security implications, and we’d rather you make it knowingly than discover it mid-rollout.
What happens once a shadow system is found?: It becomes an asset: it appears in the topology, gets scored on the 5×5 risk matrix, and shows up in posture alongside registered workloads. From there you choose the path — move it behind the gate with a base-URL change, catch it estate-wide with dns-redirect, or accept and document the risk. Discovery gives you the decision; it doesn’t make it for you.

Find the AI usage nobody registered.

Your egress logs already know which copilots and scripts are calling AI providers. Feed them to the engine and get the inventory — then put a gate in front of it.

Deploy the stack

or email hello@anchorgate.ai