Contents · 12 sections
- 011. Outside-in reviews we start ourselves
- 022. Passive reviews for an organisation a firm belongs to or works with
- 033. Authorised assessments
- 044. Checks a policyholder has agreed to
- 05What we never do, in any of the four
- 06How to identify our requests
- 07How to opt out
- 08Coordinated disclosure: exposure found in a third party
- 09Reporting a vulnerability in LeakTrace
- 10Legal basis
- 11Acknowledgments
- 12Change log
1. Outside-in reviews we start ourselves#
When we review a firm that has not asked us to, we read only what its internet presence answers to anyone who asks. Such a review can do three things and nothing more.
- It reads public records and the firm’s public pages. Public records are DNS and mail records, certificates and certificate logs, registration records, breach records and public internet indexes. It reads the firm’s website the way a browser does: its pages, the scripts they load and the standard files a site publishes for machines. Before we write to a firm it can read up to 30 of the firm’s own pages and the documents they link to, one page at a time, with the site’s robots.txt obeyed.
- It checks which services answer. It asks whether 15 commonly exposed service ports, plus two control ports, accept a connection on the firm’s website, its www host and up to six other host names under the firm’s own domain. Before we write to a firm it can ask the same on the web host of up to eight other domains the firm owns. It opens each connection, reads any greeting the service sends on its own and closes it. It reads the front page of each of those host names to see what product answers there.
- It asks whether a private file or a sign-in page answers. It sends one request to each of a short list of well-known addresses, such as
/.env,/.git/configor/wp-login.php. For a settings file, a copy of the code history or a file that holds keys, it confirms only that the file answers, from its first few bytes at most, and reads nothing more. For a memory dump it reads the answer’s headers only. For a status or API description page it reads the start of the page only to tell a real answer from an error page. It keeps nothing that is in any of them.
It never signs in, never types into or submits a form, never tries a password and never uses a weakness or anything it sees. A firm can tell us to stop at any time: email [email protected] from an address at its domain and we add the domain to our do-not-assess list within 24 hours.
The line we hold is simple. Seeing that a door is open is allowed. Walking through it is not. On a review we start ourselves we never check storage bucket names, list WordPress authors or check client-portal addresses. With an order, which is the firm’s own permission, the assessment reads further under section 3.
What we share: the result goes only to the firm itself, usually as a private briefing link sent to the firm. We do not publish it and we do not give it to anyone else.
2. Passive reviews for an organisation a firm belongs to or works with#
An association, network, adviser or other organisation may ask us for a summary across the firms it works with. Until a firm agrees on its own page (section 3), we use public records and ordinary page visits only. We make no connection checks and request no private file. We run none of the name checks in section 3.
What we share: the summary goes privately to the organisation that asked for it. It is never published.
An insurance broker or underwriter may ask us, at quote time, for the summary of one applicant that asked them for a quote and agreed to it. We make a fresh reading from public records and ordinary page visits only. We make no connection checks and request no private file. The summary goes to the broker's work address with a reference and a fingerprint anyone can check at /verify/. We keep the broker's statement that the applicant asked and agreed, with its date.
A managed IT or security provider that is our referral partner may ask us, from its partner portal, for the same summary of firms it serves or is quoting. We make the same fresh reading from public records and ordinary page visits only. The summary goes to the provider's own address. We keep its statement that it serves or is quoting each firm, with its date.
3. Authorised assessments#
A client who buys an assessment or ongoing monitoring signs an agreement that authorises it. A firm can also agree, on its own consent page, to a full check asked for by an organisation it works with. Within that agreement we run everything in section 1 against the domains named, every time, and repeat it on the schedule the agreement sets. We also run:
- Connection checks on every domain the firm owns: the same 15 service ports (for example remote desktop, database and file transfer ports) and two control ports that show whether a server answers on every port, on the website and www host of the domain ordered, up to six of its other host names and the web host of each other domain the firm owns. We open the connection, read any greeting the service sends on its own and close it. We do not log in, send credentials or try passwords.
- Name checks: whether a public storage bucket exists under names that match the firm, whether a WordPress site lists its authors publicly (we stop at three names) and whether common client-portal addresses answer.
- A deeper page read on every check: the firm’s own pages read again, the most telling first, with a visit to the common sign-in addresses. Nothing is typed or submitted and no password is tried. On monitoring it is repeated every night.
- A person’s review: every finding is verified again and checked by a person before release.
Where a sensitive file is served to the public, a paying client’s own evidence includes a short, redacted extract, so the client can see what was exposed. Any passwords, keys or tokens in that extract are removed before it is stored.
What we share: the results of a paid assessment belong to the client and go only to the client. The results of a check a firm agreed to go to the organisation that asked for it, and alerts about changes go to the firm itself, as its consent page says.
4. Checks a policyholder has agreed to#
An insurer may ask its policyholders whether they agree to a check. Nothing is checked until the policyholder agrees on its own page. The check is passive, as in section 2, and is repeated monthly while agreement stands. The policyholder can refuse, or withdraw at any time on the same link; withdrawal stops the checks and removes the results the insurer can see. We never contact a policyholder ourselves.
What we share: the results go to the insurer while the policyholder’s agreement stands.
What we never do, in any of the four#
- Use any key, password, token or other credential we find, for any purpose.
- Log in, get past a login, or reach anything that needs one.
- Change, delete or write anything on a system we review.
- Download the contents of a storage bucket.
- Try passwords, flood a service, or do anything meant to slow or break it.
- Keep, copy or use anything that is in a file we find.
How to identify our requests#
Most requests a review sends to a firm’s own website name LeakTrace in their User-Agent. A few checks send a browser’s own User-Agent, because they must see the site as a visitor’s browser does. Most identified requests carry this one:
LeakTrace-Scanner/1.0 (security audit; contact: [email protected])
A few checks send LeakTrace-EASM/2.0 or LeakTrace-Categorizer/1.0 instead.
A name that only looks like a firm’s, one somebody else registered, is read differently. Those requests carry a plain browser identity and nothing that names LeakTrace, so whoever registered the name is not told that a firm is having it looked at. They stay few: the name’s front page and a handful of its public files, with nothing typed or submitted. The identification above applies to a firm’s own domain.
How to opt out#
To ask us to stop reviewing your domain, email [email protected] from an address at that domain, or from any address that can show authority over it. We will:
- Add the domain and its subdomains to our do-not-review list within 24 hours. Our systems check that list before every review and refuse any domain on it, with no person needed to remember it.
- Stop every future review of it, of any of the four kinds above, unless the domain’s owner later signs for an assessment.
- Confirm by email when this is done.
A policyholder can also refuse or withdraw on its own consent page at any time, with no email needed.
Coordinated disclosure: exposure found in a third party#
Occasionally during a customer assessment, LeakTrace will discover an exposure that does not belong to the customer. For example, a publicly-exposed credential in a third-party vendor used by the customer, or a misconfiguration on a partner organization's domain. Our handling of these findings:
- The finding is reported confidentially to the LeakTrace customer who commissioned the assessment, the same as any other finding in their report.
- LeakTrace will not directly contact the affected third party without the customer's permission, and will not publicly disclose the finding.
- Where the customer wishes to notify the third party, LeakTrace can provide a redacted technical description suitable for forwarding, on request.
- Where the exposure represents an active, ongoing risk to the public (for example, an actively-leaking public bucket holding consumer data of a non-customer), LeakTrace reserves the right, after a 30-day customer-notification window, to disclose responsibly to the affected third party under our standard [email protected] coordinated-disclosure process, regardless of the customer's preference. We will inform the customer in writing before doing so.
- We never disclose to law enforcement, regulators, journalists, or competitors of the affected third party without an explicit legal obligation to do so.
If you are a third-party organization who believes LeakTrace has information about an exposure affecting you, email [email protected]. We will respond within 72 hours and, where appropriate, share what we can without breaching customer confidentiality.
Reporting a vulnerability in LeakTrace#
If you have discovered a security vulnerability in any LeakTrace property (getleaktrace.com or any subdomain or product surface), please report it to [email protected]. We commit to:
- Acknowledge your report within 72 hours
- Provide a status update within 7 days
- Resolve confirmed critical issues within 30 days
- Credit you publicly (if you wish) on the Acknowledgments section below
We do not currently operate a paid bug-bounty program. We deeply appreciate responsibly-disclosed reports.
Legal basis#
Our reviews rely on information that the organisation already serves to the public, and on the written agreement or recorded consent where section 3 or 4 applies. We do not seek access beyond what is served publicly, and we do not use anything we find to gain further access. If you have a question about a specific review, contact us and we will provide the dates, the requests we made and our method.
Acknowledgments#
We thank all security researchers who have responsibly disclosed issues to us. (No public acknowledgments at this time.)
Change log#
- 20 November 2026 (version 2026-11-20.1): the documents now say one thing about a review we start ourselves. Section 1 says it in the same words as the Privacy Policy (section 4) and the Authorized Scan Agreement (section 3): it reads public records and the firm’s public pages; it checks which services answer on the firm’s own host names; it asks whether a private file or a sign-in page answers at a short list of well-known addresses and keeps nothing from it. It never signs in, submits a form, tries a password or uses a weakness. Section 3 says what an order adds. The entry of 30 September 2026 below described the reviews of that date.
- 17 November 2026 (version 2026-11-17.1): section 1 now lists the three things a review we start ourselves does. It reads public records and public pages. It sees which services answer and what product they are. It sees whether a private file answers and keeps nothing from it. Before we write to a firm we also read up to 30 of its own pages and its other domains. We still never type into a form, try a sign-in or use what we see. Sections 2 and 3 say what the other kinds of review leave out and what an assessment adds.
- 22 October 2026: section 6 says how a name that only looks like a firm’s is read: with a plain browser identity, so whoever registered the name is not told that a firm is having it looked at. Requests to a firm’s own domain still name LeakTrace.
- 18 October 2026: section 3 adds the deeper page read an authorised assessment and monitoring run. A review we start ourselves still does not run it.
- 30 September 2026: a review we start ourselves no longer requests hidden paths or files such as
/.envor/.git/config; that check now runs only on an authorised assessment. Section 1 no longer lists connection checks, which have not run on reviews we start ourselves since 29 September 2026. Section 3 now lists every check an authorised assessment adds. The User-Agent section names every identifier our reviews use. - 30 September 2026, earlier: product name updated to External Cybersecurity Assessment. Scope and limits unchanged.
- 29 September 2026: security reports go to [email protected].
| Date | What changed | Applies to |
|---|---|---|
| 20 Nov 2026 | Version 2026-11-20.1. Section 1 now says what a review we start ourselves does in the same words as the Privacy Policy and the Authorized Scan Agreement. Section 3 says what an order adds. Full detail in the change log above. | From 20 Nov 2026 |
| 17 Nov 2026 | Version 2026-11-17.1. Section 1: a review we start ourselves reads public records and public pages, sees which services answer and what product they are and sees whether a private file answers without keeping anything from it. Before we write to a firm we also read up to 30 of its pages and its other domains. Sections 2, 3 and 5 follow. Full detail in the change log above. | From 17 Nov 2026 |
| 22 Oct 2026 | Section 6: a name that only looks like a firm’s is read with a plain browser identity, so whoever registered it is not told that a firm is having it looked at. Requests to a firm’s own domain still name LeakTrace. | From 22 Oct 2026 |
| 30 Sep 2026 | Hidden-path requests (/.env, /.git/config, key files) are now authorised-only. Section 1 corrected: no connection checks on a review we start ourselves. Section 3 lists every check an authorised assessment adds. Full detail in the change log above. | From 30 Sep 2026 |
| 30 Sep 2026 | Product name updated to External Cybersecurity Assessment. Scope and limits unchanged. | From 30 Sep 2026 |
| 29 Sep 2026 | Security reports go to [email protected]. | From 29 Sep 2026 |