Methodology

Why hostname evidence beats every other detection method

Surveys lie, app inventories age, agents miss the browser. Your network logs already record every AI tool anyone reached. This page explains how we turn that record into a defensible inventory.

20,399classified AI tool domains, maintained daily
~300,000new domains checked every day
120Mdomain corpus screened for AI signals
18functional categories in the register
The premise

Every AI tool leaves the same fingerprint

Whatever the tool does, it resolves a hostname. That is the one signal no AI product can avoid emitting.

It cannot be faked away

An employee can skip a survey or shrug at an interview. Their browser still resolved chatgpt.com at 9:14.

It is already historical

Your logs hold weeks or months of the past. Detection starts with a backlog, not from zero on install day.

It costs nothing to collect

The DNS filter, proxy or firewall you already run is the sensor. There is nothing to deploy and nothing to maintain.

The method in one sentence: take the hostnames your network already logged, and match them against a live register of 20,399 classified AI tool domains.

The alternatives

What the other methods miss

Each alternative answers a different question. None of them answers "what did our people actually reach".

Surveys and self-reporting Weakest

  • People underreport tools they suspect are not allowed. Not from malice, from caution.
  • Nobody remembers the AI feature embedded inside a SaaS product they think of as "the CRM".
  • Results age the day they are collected.

App inventories and CASB catalogs Partial

  • Strong on sanctioned SaaS, weak on the long tail. New AI tools appear faster than catalogs classify them.
  • A tool used through a personal account on the web often never appears as an "app" at all.

Endpoint agents Heavy

  • Deployment is a project: procurement, rollout, exceptions, performance complaints.
  • Unmanaged and BYOD devices stay invisible, and those are where shadow use concentrates.

Log-based hostname audit This method

  • Covers every device that touches the logged network, managed or not.
  • Works retroactively on history you already have.
  • Its accuracy is decided by one thing: the quality of the register it matches against.

A fuller comparison, including when each method genuinely wins, is on the detection methods page.

The register

Where the 20,399 domains come from

The match is only as good as the list. The register is a maintained commercial product, not a scraped blocklist.

CorpusA 120 million domain corpus provides the candidate stream.
Daily screenRoughly 300,000 new domains are checked every day for AI signals.
ClassificationConfirmed tools are placed in 18 functional categories with subcategories.
Terms reviewVendor terms are read for training verdicts, 13,000+ reviewed to date.
FlagsRisk level, data sovereignty and abusive-purpose flags complete each record.

Training verdicts

Our analysis finds 700+ tools that train on customer data by default. Each verdict is dated to when the terms were checked.

The silence finding

85.5% of AI tools say nothing about training in their public terms. Silence is recorded as its own verdict, never guessed over.

Provider mapping

3,900+ tools are mapped to the model provider actually behind them. A rebranded wrapper inherits the risk of its backend.

Matching mechanics

Subdomain walk-up, explained

Logs record hostnames at every depth. The matcher starts specific and walks up until it finds the registered tool.

query: api.eu.example-ai.com try: api.eu.example-ai.com no entry try: eu.example-ai.com no entry match: example-ai.com AI tool, risk: medium query: gemini.google.com match: gemini.google.com exact, AI-native note: google.com alone would NOT match here

Two real behaviors: walking up to the registered domain, and stopping at an exact subdomain so a whole platform is not condemned for one AI feature.

Why walk-up matters

AI tools scatter their traffic across api, cdn, auth and regional subdomains. Without walk-up, an audit undercounts by whole tools.

Why exactness matters more

The reverse mistake is worse: matching google.com because gemini.google.com is AI would flag all of Google. Depth-aware entries prevent that.

URLs become hostnames

Proxy logs carry full URLs. The parser reduces them to hostnames before matching, so paths and query strings never influence a verdict.

Signal handling

From raw lines to defensible counts

Between your upload and the report, five things happen to the data. All five are visible in the output.

StageWhat happensWhere you see it
Format detectionCSV headers, key=value syslog or plain lists are recognized automatically.The scope line names the detected format.
ParsingHostname, identity and timestamp fields are extracted. Unparseable lines are skipped and counted.Lines read vs lines parsed in the report metadata.
DeduplicationRepeated queries collapse into hit counts per domain and per identity.The hits and users columns.
MatchingEvery distinct hostname runs the walk-up match against the register.The tool table, plus a count of unmatched domains.
EnrichmentMatched tools pull category, risk, sovereignty, abusive and training fields, plus your sanctioned list.Every column right of the domain.

The uploaded export is processed in memory and discarded after the run. Reports live 90 days in your account and can be deleted earlier.

Failure modes

Where this method has limits, honestly

A methodology page that lists no limits is a marketing page. These are ours, and how to work around each.

Encrypted DNS to third parties

Devices using DoH to an outside resolver bypass your DNS logs. Proxy and firewall logs still catch the connection, which is one reason to audit more than one source.

Off-network use

Home Wi-Fi and cellular data never touch your logs. The audit measures the network you log, and states so in the evidence statement.

Brand-new domains

A tool launched yesterday may not be classified yet. The daily screening loop and the unmatched-domain count keep this window short and visible.

Shared platform domains

AI features living entirely on a suite's main domain are hard to separate from normal suite use. Depth-aware entries catch the ones with distinct hostnames.

Identity gaps

No identity column means no per-user table. The tool inventory is unaffected.

It observes, it does not enforce

The audit produces evidence. Blocking, if you choose it, stays in your own stack. Sequencing advice: detect before you block.

See the withheld-rows preview for yourself

The sample evidence pack shows exactly what this methodology produces: tiles, verdicts, the tool table and the per-user breakdown, on sample data.

Open the sample PDF
Comparisons that matter

Method properties, side by side

When someone proposes an alternative in your review meeting, this is the table to have open.

PropertySurveyCASB catalogEndpoint agentLog audit
Covers BYOD on your networkNoPartlyNoYes
Sees historical usageMemory onlyPartlyFrom install dayYes, full log window
Deployment effortLowHighHighNone
Long-tail AI tool coveragePoorLagsDepends on catalogRegister updated daily
Training-terms contextNoRarelyNoDated verdict per tool
Cost to tryMeeting timeProcurementProcurementFree preview

These methods also combine well. Teams with a CASB still run log audits, because the audit catches what the catalog has not classified yet and provides the dated training verdicts.

Taxonomy

The 18 functional categories

Every matched tool lands in exactly one category. Categories map findings to the department that owns the conversation.

Creation

  • Text & Language: assistants, writing, translation.
  • Image & Visual: generation, editing, upscaling.
  • Video & Animation: generation, avatars, editing.
  • Audio, Voice & Music: speech, voice cloning, transcription.
  • Design & 3D: interfaces, models, rendering.

Work

  • Code & Development: assistants, agents, review.
  • Productivity & Collab: notes, documents, meetings.
  • Search, Knowledge & Docs: answer engines, research.
  • Data & Analytics: analysis copilots, BI.
  • Marketing & Sales: content, outreach, ads.
  • Agents & Automation: autonomous agents, workflows.

Sector and infrastructure

  • Education & Learning, Health & Wellbeing, Finance & Legal.
  • Entertainment & Companions, Robotics & Embodied AI.
  • Models & Infrastructure: hosting, APIs, model hubs.
  • Security & Moderation: detection, filtering, redaction.

Subcategories go a level deeper, over a hundred of them, which is what lets policy profiles treat "browser agents" differently from "RPA with AI".

Worked example

One export, three very different verdicts

Three hostnames from the same sample export show why per-tool context beats a yes/no AI flag.

claude.ai

  • AI-native assistant, risk low, does not train on user data per checked terms.
  • Verdict under the General business profile: allow.
  • The audit's job here is confirmation, not alarm.

otter.ai

  • Meeting transcription, so it hears whole conversations. Trains unless opted out.
  • Verdict: allow with controls, meaning enterprise tier and the opt-out flipped.
  • Without the training verdict, this tool looks harmless.

deepswap.ai

  • Face-swap generation, flagged abusive-purpose, trains by default.
  • Verdict: block, and the per-user table shows exactly one source reached it.
  • That single row justifies the whole audit on its own.

All three of these appear in the sample report, so you can see how the verdicts render in the actual document.

Scrutiny

When your auditor questions the method

Auditors probe methods, and this one holds up. Map their questions straight to report sections.

The auditor asksThe method's answerWhere it is written
"What was the data source?"The organization's own DNS, proxy or firewall export.Scope line and evidence statement.
"What period does this cover?"The export window you chose, stated with line counts.Report metadata.
"What was it matched against?"A register of 20,399 classified AI domains, maintained daily.Scope line, with the register size at run time.
"How current are the vendor claims?"Every training verdict carries the date the terms were checked.Tool table, per row.
"Is it repeatable?"Same window, same source, re-run any time. Differences between runs are findings.Evidence statement.
"What are the known blind spots?"Off-network use and third-party DoH, stated rather than hidden.This page, and your filing notes.
The leftover column

What unmatched domains tell you

Not every hostname in your export is an AI tool, and the report says how many were not. That number is useful twice.

As a sanity check

  • A normal export is mostly non-AI traffic. If nearly everything matched, your export was pre-filtered.
  • Compare unmatched counts between runs to confirm your export settings stayed constant.

As an early-warning list

  • A domain that is unmatched today can be classified tomorrow, since screening runs daily.
  • Re-running last month's window occasionally catches tools that were too new the first time.
FAQ

Methodology questions

How accurate is hostname-based detection?

For "was this tool reached from this network", it is as accurate as your logs. The register's daily maintenance and depth-aware matching keep false positives down.

Can it tell who used a tool?

When the export carries a user, device or IP column, yes, per identity. Without one you still get the full tool inventory.

What about tools accessed through APIs?

API traffic resolves hostnames too, and API subdomains are in the register. Server-to-server AI use shows up like any other source.

How fresh is the register?

It is maintained daily, with roughly 300,000 new domains checked each day from a 120 million domain corpus.

Does a match mean data left our network?

It means the hostname was resolved or the connection was made. Hits and user counts tell you whether it was a one-off lookup or sustained use.

Why not just use our firewall's AI category?

Built-in categories are coarse and update slowly. The audit adds per-tool risk, training verdicts with dates, and the sanctioned split, which category filters cannot express.

Build vs buy

Why teams stop maintaining their own AI blocklists

Many IT teams start with a hand-kept list of AI domains. The math turns against them within a quarter.

The arrival rate

  • New AI tools launch daily, and existing ones add domains and subdomains constantly.
  • A list that was complete in January quietly stops being complete by March.

The context problem

  • A domain name alone says nothing about training terms, sovereignty or abusive purpose.
  • Reading vendor terms at scale is its own job, 13,000+ reviews and counting in our register.

The audit gap

  • A homegrown list produces a block decision, not evidence. There is no dated verdict to file.
  • Consulting-led shadow AI assessments typically take weeks and are quoted in the thousands. The audit compresses that into an upload.

The practical split: let a maintained register carry the classification burden, and spend your team's time on the decisions the report tees up. Costs are on the pricing page.

Test the methodology on your own logs

The free preview applies the full pipeline to your export and shows the totals. Judge the method by its output.

Run the free audit