Database credential
Backup scripts, ORM configs, Terraform state, legacy services
What it catchesDatabase access through leaked connection strings or compromised backup archives.
THE PLATFORM
Two layers. One workspace.
Built for the company without a security team.
Decoys catch the intrusion the moment it starts, and Vantuz investigates and responds while you read the alert. The Exposure Sonar closes the doors before anyone tries them. Both written for the person who runs the company, not for a SOC.
11decoy formats, from AWS keys to browser sessions
<60sfrom touch to evidence dossier
0agents to install, nothing changes on your network
EUdata plane, UK and EU GDPR
01 · PLANT
Put a decoy where
an attacker would look.
Decoy credentials, documents, packages, AI knowledge-base files and browser sessions, in the formats attackers already hunt for, placed in minutes where your team already works. No agent, no change to your network.
See alsoEvery decoy format
02 · KNOW
A touch is the start
of the story, not the end.
Source, network, intent, technique and a risk score you can read come together in one place, with a plain-English write-up. Human, script or AI agent, Vantuz tells you which.
See alsoThe investigation pipeline
RESPONSE
When a decoy fires,
this is what answers.
Vantuz always alerts you. Connect a provider and it can also act, within limits you set: challenge or block the source, or send the event to your SIEM. Every action is logged and reversible, and containment is off until you turn it on.
Challenges or blocks the source at the edge, before it reaches your servers.
Adds the source to a WAF rule in your own AWS account. It counts first and blocks only when you allow it.
Sends a signed event to your SIEM, Slack or anything with an HTTPS endpoint.
Cloudflare · counting only
EXPOSURE SONAR
Close the doors
before anyone tries them.
Most breaches start with something left open: a config file on the web, an admin without two-step verification, email anyone can forge. The Sonar finds it from the outside, explains the fix in plain steps, and confirms when it is really closed.
Read-only checks on the domains and accounts you proved are yours: email that can be forged, files and code exposed, databases reachable from the internet, admin accounts without two-step verification.
Every finding comes with what it is, why it matters, the risk of changing it, the steps, a way back, and a message you can hand to your IT provider. It is marked fixed only when a later check confirms it.
Where you cannot fix it yet, place a decoy right in the exposed spot. It alerts you the moment anyone tries to use it.
Findings that need action4 open
Configuration file with secrets exposedstaging.example.com/.env
Email from your domain can be forgedexample.com
Source code history exposed (.git folder)shop.example.com
Admin account without two-step verificationGoogle Workspace
Last checked today, 09:41Interface concept · Sample data
DECOY FORMATS
Eleven formats. Each one imitates something an attacker wants.
A decoy only works if it looks like the real thing. Each format is built to pass for the credential, file, package or login an intruder is hunting for, and to report when it is used. Where a format has a limit, its card says so.
Where it lives
Document tracker
Shared drives, M&A folders, board decks, HR PDFs
What it catchesDocument theft and reading by people who should not have it.
URL beacon
Internal wikis, Notion pages, pinned messages, admin docs
What it catchesLateral movement and phishing landings through collaboration tools.
DNS hostname
VPN and SSH configs, runbooks, hosts files, internal docs
What it catchesReconnaissance: anyone who finds the hostname in your files and tries to reach it.
RAG document
Vector databases, LLM knowledge bases, internal AI assistants
What it catchesAI agents and LLMs pulling internal documents out through RAG pipelines.
npm package
A decoy package under your org namespace or a plausible typosquat
What it catchesDependency confusion and typosquatting attempts.
Browser session
Early accessOne fake login session, kept by a small extension in Chrome or Edge
What it catchesInfostealer malware and hostile extensions that copy the browser's cookies, when someone replays the decoy session.
How the browser decoy worksAWS credential
Env files, ~/.aws/credentials, backup folders, anywhere a cloud key would sit
What it catchesSomeone using the key. It is a real AWS access key that is not allowed to do anything: AWS answers only the identity check and refuses every other call. You get an alert, usually within a few minutes and often within seconds.
Honest limitReading or copying the key does not fire it; using it does. Every decoy key lives in one Vantuz AWS account, so someone who has already seen one could recognise another without using it.
CI/CD secret
A pair of GitHub Actions repository secrets
What it catchesSomeone using the key after taking it from your pipeline. The secrets hold a real AWS access key that is not allowed to do anything, and you get an alert, usually within a few minutes and often within seconds.
Honest limitReading or copying the secrets does not fire it; using the key does. Every decoy key lives in one Vantuz AWS account, so someone who has already seen one could recognise another without using it.
OpenAI key
Agent configs, .env files and internal tooling, as a key plus an API address on your own verified domain
What it catchesA tool using the planted configuration. The tool gets the provider's ordinary "out of credit" answer, and your alert names the tool, the model it asked for, the country and the network.
Honest limitA key copied and used directly at OpenAI is not seen.
Anthropic key
Agent configs, .env files and internal tooling, as a key plus an API address on your own verified domain
What it catchesA tool using the planted configuration. The tool gets the provider's ordinary "out of credit" answer, and your alert names the tool, the model it asked for, the country and the network.
Honest limitA key copied and used directly at Anthropic is not seen.
BROWSER DECOY · EARLY ACCESS
Stolen sessions,
caught when used.
Infostealer malware and hostile extensions copy the cookies that keep you signed in, then replay them from their own machine. The browser decoy keeps one fake session next to the real ones. If anyone replays it, they get an ordinary 404 and you get an incident with the source and what to do first.
See how the browser decoy worksIllustrative replay · sample data
- 01A decoy sessionThe extension keeps one fake login session in the browser, next to the real ones.
- 02The jar is copiedInfostealer malware copies every session in the browser, the decoy included.
- 03Replayed, answered 404The attacker replays them. The decoy domain answers with a plain 404, so they learn nothing.
- 04You knowVantuz opens an incident, scores it High and emails you plain next steps, starting with a quick check with the person.
THE FIRST MINUTE
From touch to evidence in under a minute.
Deterministic where it matters, explained everywhere. AI writes the investigation and recommends the hardening; it never makes a destructive decision on its own.
Decoy touched
A decoy credential, URL, document, package or browser session reports activity.
Source enriched
Geolocation, network, reputation. Only the fields needed to judge the source.
Risk scored
Deterministic 0 to 100. The same inputs always give the same score, and you can see why.
MITRE ATT&CK mapped
Tactic and technique attached, ready for your SIEM or your report.
Actor classified
Human, script or autonomous agent, with the evidence behind the call.
Response prepared
Recommended hardening, suggested follow-on decoys and the containment options you enabled.
Evidence dossier ready
An incident PDF with timeline, evidence and response notes for security and compliance review.
WHERE THE NEXT MOVE HAPPENS
Your environment.
Your decoys.
From a repository to an AI knowledge base, give an intruder something that gives them away. Three of the ways companies use Vantuz today.
A leaked database password, a stolen API token, a cloned repository. Decoy credentials next to the real ones turn the first use into a signal, before the attacker reaches production.
Ransomware crews and curious insiders read before they act. A decoy contract, board deck or HR file tells you the moment someone opens what they should not.
Your assistants read your knowledge base; so does a hostile agent. Decoy documents in the knowledge base catch an agent, yours or theirs, the moment it goes where it should not. A decoy OpenAI or Anthropic key, planted with an API address on your own verified domain, fires when a tool uses that configuration.
SEE IT LIVE
Watch it catch an attack.
Then decide.
Thirty minutes, your environment. We plant a decoy, trigger it on purpose,
and you watch the whole response happen. No slides.
Do I need to install anything?
No. Vantuz is agentless. You place decoys by hand or through the API, and the Exposure Sonar checks your domains from the outside, read-only. The one optional exception is the browser decoy, in early access: a small Chrome or Edge extension that holds a single fake session and cannot read your real cookies or browsing.
What does the Sonar actually do to my systems?
It sends read-only requests to hosts under domains you have proven you control, at a deliberately slow rate. It never attempts to exploit anything, and it never reads your mail, files or chat.
Can Vantuz block things on its own?
Only if you connect a provider and enable it. Containment is off by default, every action is logged and reversible, and the destructive ones run on rules you set, never on an AI's judgement.
How is a finding marked as fixed?
By a later check confirming it from the outside, never by a click. If the check cannot see, the finding stays open and the coverage report says so.
How do the AWS credential and CI/CD secret decoys work?
Both rest on a real AWS access key that is not allowed to do anything: AWS answers only the identity check and refuses every other call. The AWS credential is that key; the CI/CD secret is a pair of GitHub Actions repository secrets holding such a key. When someone uses the key, the use is recorded by AWS itself, Vantuz is told and you get an alert, usually within a few minutes and often within seconds. In our live tests Vantuz saw the use as soon as 3 seconds after it happened, and the alert email arrived about 21 to 28 seconds after use; that is what we measured, not a guaranteed time. Reading or copying the key does not fire it; using it does. One honest limit: every decoy key lives in one Vantuz AWS account, so someone who has already seen one could recognise another without using it.
How do the OpenAI key and Anthropic key decoys work?
Each one is an AI configuration decoy: a key plus an API address on your own verified domain. It fires when a tool uses the planted configuration. The tool gets the provider's ordinary "out of credit" answer, and you get an alert naming the tool, the model it asked for, the country and the network. In our live tests the alert email arrived about 20 seconds after use; that is what we measured, not a guaranteed time. One honest limit: a key copied and used directly at OpenAI or Anthropic is not seen.