Skip to main content

Threat Report: Security Tools That Publish Your Private Links

Monadnock Cyber

Threat Report: Security Tools That Publish Your Private Links

Nov 13, 2026 · @Jeff Stutzman

Paper 3 of 14 on systemic architectural risk in standard security tooling, not on specific vendors, unpatched vulnerabilities or any one organization's configuration.

Bottom line

Most email security tools check the links in the messages you receive before you click them. Some of it does that by handing the link to a public website that visits it, takes a picture of the page, and keeps the result where other people can search it. For a phishing link, that is the point. For a password-reset link, a DocuSign envelope or a shared OneDrive file, it means your security stack just published a key to your account or your document. Nobody at your company sees it happen, and the person who sent you the link never finds out.

Researchers have documented this at least five times since 2022. Each time a vendor or a setting got fixed, and each time the pattern came back somewhere else.

Three ordinary things combine to produce this risk. None of them is a vulnerability on its own, but together...

1. Links that work as keys. 

  • A link used to point at a page; now it often is the permission. 
  • A password-reset link signs you in long enough to change the password. 
  • An "anyone with the link" share opens the file with no sign-in. 
  • A DocuSign link opens the contract. 
Developers call these capability URLs: the long random string in the link does the job a password used to do. They stay safe exactly as long as nobody else sees the link.

2. Public URL scanners. Services such as urlscan.io, VirusTotal and Hybrid Analysis take a link, open it in an isolated browser, record what loads (screenshot, scripts, redirects) and judge whether it looks malicious. Security analysts use them every day, for good reason. Each scan carries a visibility setting:

  • Public: anyone can search it.
  • Unlisted: hidden from the public search, but visible to the service's paying subscribers.
  • Private: visible only to the account that submitted it.

3. Automation that submits every link. Security teams automate the routine work. A SOAR platform (security orchestration, automation and response) runs a playbook on each arriving email: pull out every link, check each one against a scanner, act on the verdict. Mail gateways, endpoint tools and browser extensions do the same. When the integration submits with public or unlisted visibility, the playbook publishes every link it checks, including the ones that are keys.

Put the three together and the tool meant to protect the recipient becomes the leak.

How an attacker uses it

Positive Security, which documented the problem in 2022, laid out the account takeover:

  1. Find people whose tools auto-submit. Scrape the scanner's public results for email addresses. Addresses that keep turning up in scanned reset, unsubscribe and invitation pages belong to people whose mail pipeline submits links.
  2. Trigger a reset. Request password resets for those addresses at common web services.
  3. Let the pipeline publish it. The victim's security stack receives the reset email and submits the link to the scanner.
  4. Collect it. Search the scanner for recent scans of that service's reset pages and use the link before the user does.

No malware, no phishing and no break-in is involved. The attacker reads what the victim's own tools put on display.

Key judgments

  1. Very likely, high confidence: Security integrations still publish inbound private links today. Independent and academic researchers found it separately between 2022 and 2026, the latest in May 2026. Misconfigured SOAR playbooks and email security integrations submit inbound links to public scanners, and password-reset links, DocuSign requests, shared-file links and invoices sit in searchable scan results as a result.
  2. High confidence: Every documented case traces to a visibility setting. In all but one case the setting was a configuration choice; the one product default, in Palo Alto XSOAR's urlscan.io integration, was fixed in May 2022. 
  3. High confidence: Leaked links in the research opened accounts, documents and meetings with no addition login or action required. A leaked link is worse than a leaked file: the file is one copy of one document, while the link is a working key until it expires.
  4. Likely, moderate confidence. An attacker finds people whose mail is auto-submitted by reading the scanner's results. Positive Security emailed 47 test links to organizations; 21 appeared in public scans.
  5. Moderate confidence: One researcher's 2024 finding and urlscan.io's documentation support it. "Unlisted" does not mean private. Unlisted scans stay searchable by the scanner's paying subscribers, and at least one integration forces that setting even when the account is set to private.

What a published link gives away

The link types researchers have found in public scan results and URL archives, and what each grants whoever finds it:

Link type Examples found What finding it grants
Password reset and account setup Instagram reset confirmation, Zendesk account setup, Apple ID finalization Takeover of the account, if the link is still valid
Document signing DocuSign signing requests for contracts The contract's terms and parties; sometimes the ability to sign
Shared files Google Drive, Dropbox, SharePoint (including a government invite), OneDrive, iCloud The file, with no login: tax records, invoices, medical documents
Meetings and recordings Government Zoom invites including court hearings, Webex recordings, Discord invites Entry to the meeting or the recording
Payments PayPal invoices and money claims, accounting-platform invoices Amounts, counterparties, sometimes payment action
Embedded secrets A live VirusTotal API key, Telegram bot tokens Use of the service as the key's owner
Login flows Microsoft sign-in URLs carrying authentication redirect data (found in the Wayback Machine archive) OAuth replay, if the redirect data is still valid
Unsubscribe and tracking Unsubscribe pages showing full email addresses, DHL tracking Identity confirmation for targeting

Sources: Positive Security, 2022; Pulse Security, Jan 2026.

At scale:

  • A February 2026 study of 6.1 million URLs from VirusTotal, urlscan.io, Hybrid Analysis, the Wayback Machine and RedHunt found 12,331 leakage incidents, by the authors' count, covering authentication, financial, personal and document data (Ramadan et al., arXiv 2602.21826).
  • A May 2026 study of 332,000 URLs from the public feeds of six URL-scanning services found more than 4,000 that leak sensitive personal information, at 97% precision. Honeypage links the researchers submitted were opened by outside parties (Mustafa et al., LEAKYLINKS, IEEE Symposium on Security and Privacy 2026).

Documented cases

Each fix addressed one vendor or one setting. The pattern kept coming back.

Date Case What it showed Source
May 2026 LEAKYLINKS study (CISPA) More than 4,000 URLs leaking sensitive personal information in the public feeds of six URL-scanning services; submitted honeypage links opened by outside parties IEEE S&P 2026
Feb 2026 Silent Spill study 12,331 leakage incidents across 6.1 million URLs from scanning platforms, paste sites and web archives arXiv 2602.21826
Jan 2026 Pulse Security review of "unguessable" links Dropbox, Google Photos and OneDrive share links in urlscan.io and AlienVault OTX, including medical documents; some endpoint products submit URLs automatically Pulse Security
Mar 2024 Independent researcher (vin01) Private file-share links on urlscan.io, Hybrid Analysis and Cloudflare's URL scanner; TheHive's Cortex analyzer forced scans to "unlisted" even when the account was set to private; canary links were opened within an hour GIGAZINE summary; OWASP Frankfurt slides
Nov 2022 Positive Security research Searchable reset links, DocuSign requests, file shares and invoices; 21 of 47 test emails triggered public scans; one employer traced a leak to its own SOAR playbook Positive Security
May 2022 Palo Alto XSOAR integration bug A visibility argument defaulted to public and overrode other settings; fixed May 1, 2022 Positive Security
Feb 2022 GitHub notifies customers An automated GitHub process sent GitHub Pages sites from private repositories to urlscan.io, exposing repository names Positive Security

Who submits. The research names five SOAR integrations: Palo Alto XSOAR, Swimlane, IBM Resilient (QRadar SOAR), Splunk Phantom (Splunk SOAR) and Dimension Data's SOAR. urlscan.io listed 26 commercial products and 22 open-source projects that integrate its API. More than half of the scans Positive Security studied used a generic python-requests user agent, so the scans do not identify most submitters.

The scanner's response. In July 2022 urlscan.io added a team-wide maximum-visibility setting, published Scan Visibility Best Practices, contacted its heaviest public submitters and began deleting scans that match sensitive patterns. Those are real fixes. They work only for the accounts that turn them on.

Original test

The certificate trap measured how fast scanners act on a public certificate. This test does the same for links: unique canary links sent into mail pipelines we control or have permission to test, with every fetch and every public appearance recorded. Findings pending; the method is ready to run.

Method

  1. Mailboxes. Test mailboxes on Monadnock Cyber's own domains, plus client mailboxes only with written consent. Cover the common protection stacks: Microsoft Defender for Office 365, Check Point, Google Workspace, and at least one SOAR playbook with a URL-scanner integration.
  2. Canary links. Links unique per mailbox and per link type, on a domain used only for this test, in three shapes: a fake password reset, a fake file share and a fake signing request.
  3. Collection.
    • Log every fetch of every link: time, source address, network, user agent.
    • Search urlscan.io, VirusTotal and Hybrid Analysis daily for the test domain.
    • Record any link that appears in public results, and how soon.
  4. Metrics:
    • Share of protection stacks whose links appear in any public result
    • Time from delivery to first fetch, and to first public appearance
    • Number of distinct fetchers per link, and which are not the mail vendor
    • Fetches after public appearance, from addresses outside the vendor's network
  5. Duration. 15 days, matching the certificate trap.

Rules

  • No canary links go to third parties without consent. Positive Security's method of emailing reset and unsubscribe addresses at other organizations is not repeated here.
  • Links lead to a static page with no data. Nothing is collected from anyone who opens one beyond the request itself.
  • Results are reported by protection stack and configuration, not by client.

Impact on the defense industrial base

Defense contractors move Controlled Unclassified Information (CUI) through exactly the link types that leak: SharePoint and OneDrive shares, signing requests, and file-transfer links between primes and subcontractors.

  • CUI by link. A share link to a CUI document that lands in a public or unlisted scan exposes the document, not just its name. That is an uncontrolled flow under NIST SP 800-171 3.1.3 and a connection to an unlisted external system under 3.1.20. If someone outside opens the document, it is likely reportable under DFARS 252.204-7012. Reporting is a question for counsel.
  • The leak happens on the receiving end. The link escapes through the recipient's security stack, not the sender's. A prime that shares a CUI link with a subcontractor whose MSSP auto-submits URLs has exposed CUI through a third party it cannot see.
  • Direct account takeover. A contractor whose mail pipeline publishes links is open to the attack above: request a password reset for a cloud service, then search the scanner for the link.
  • A second effect: consumed links. Mail security that clicks every link to inspect it also uses up single-use reset and sign-in links before the user sees them. Users request the link again and learn to distrust the security tooling. That is a reliability problem, not an exposure, but it explains why organizations route around the controls.

How to tell if it is happening to you

Three checks. Seeing unlisted results takes a paid scanner account.

  1. Search the scanners for your own domains. Search urlscan.io, VirusTotal and Hybrid Analysis for your domains and your file-sharing tenant (for example yourcompany.sharepoint.com or yourcompany-my.sharepoint.com).
    • Public results need no account; unlisted results need urlscan Pro.
    • Any hit containing a share, reset or signing link is an exposure.
  2. Audit every integration that submits URLs. For each SOAR playbook, mail gateway, EDR tool, browser extension and threat-intelligence platform, record:
    • which scanners it calls
    • the visibility it requests (public, unlisted or private)
    • whether the account enforces a maximum visibility
  3. Plant a canary. Email yourself a unique link on a domain you control and watch for fetches and public appearances. That is the original test above, run on one mailbox.

What to do about it

The fixes split between the organization receiving links and the organization issuing them.

If your tools scan links

  • ☐ Private by default, enforced. Set every scanner account to private and turn on the account-wide maximum-visibility setting, so no integration overrides it.
  • ☐ Audit integrations. Check each playbook and connector for a hard-coded public or unlisted setting, and re-check after every upgrade.
  • ☐ Scan inside, not outside. Use the mail vendor's own link detonation or an internal sandbox for inbound mail; keep public scanners for links an analyst chooses to check.
  • ☐ Exclude your own tenants. Never submit links on your own domains or file-sharing tenants to outside scanners.
  • ☐ Clean up. Report sensitive results for removal, ask scanner operators to exclude your domains, and rotate any exposed keys or links.
  • ☐ Flow it down. Require MSSPs and subcontractors who handle your links to follow the same rules, by contract.

If your systems issue links

  • ☐ Don't let the link be the key. Require sign-in or an emailed one-time code before a shared file opens.
  • ☐ Expire and limit. Short lifetimes and view limits on share, reset and signing links.
  • ☐ Require a human action. Put a "continue" button between the link and the action, so a scanner's visit neither exposes content nor uses up the link.

Intelligence gaps

These gaps hold judgments 4 and 5 at moderate confidence. The original test closes the first two for the stacks it covers.

  1. Prevalence. No public measure of how many organizations, or defense contractors, run integrations that submit links publicly today.
  2. Current vendor defaults. Integration defaults change with each release; the 2022 and 2024 findings do not establish how 2026 versions behave.
  3. Who reads unlisted results. Scanner operators vet their paying subscribers, but they do not publish the subscriber base or its access logs.
  4. Exploitation. No public report of a confirmed account takeover or CUI loss through a scanned link. The LEAKYLINKS honeypages show outside parties opening submitted links, but not what they did next. The harm is demonstrated in concept and in exposed data, not in an attributed attack.

Sources

Are you at risk? Contact Monadnock Cyber to find out. 

TLP:CLEAR · © 2026 Monadnock Cyber, LLC. All rights reserved.

Comments

Popular posts from this blog

Daily Cyber Brief · 17 October 2026 · a Monadnock Cyber summary

Daily Cyber Brief 17 October 2026 · a Monadnock Cyber summary THE BOTTOM LINE N0n targets transportation firms with shipping data. Edge device flaws remain actively exploited by attackers. Supply chain compromise affects Microsoft 365 tenants. WHAT IT MEANS New Gang Targets Transportation Sector With Shipping Data The N0n group posted 38 victims in one batch on October 12, all sharing a deadline of October 19. Thirty-five of these victims are transportation organizations across 18 countries. The stolen data consists of electronic shipping messages (EDI archives). This is likely one campaign, not separate attacks. If the gang publishes the files as scheduled, the sector faces public exposure. We do not know how the gang entered these systems. One entry point these victims share, such as a common software supplier or IT provider, fits the data; so does one gang working through a target list with the same method. The evidence does not separate the two: roughly even chance either way, low...

Cyber Intelligence (Risk) Brief — a sample for your review

What if one of these arrived on a regular schedule? This is a sample Intelligence Brief from Monadnock Cyber. Corvane Precision,  its people, domains and vendors  are all fictitious. The threats are real as of October 19, 2026: the exploited flaws, the advisories, the regulatory picture. The brief answers four questions: What is coming for this organization? Why? What to watch for. What to do about it.  Every finding names its source and says how reliable that source is. Every judgment says how confident we are and why. Now picture one written about your own company, your organization, or you, delivered every quarter or whenever something changes. Monadnock Cyber offers it for: Companies: your exposure, the attackers interested in your sector, and the regulations that apply when something goes wrong. Organizations: nonprofits, schools, municipalities and associations, which often hold sensitive data with thin security staff. Individuals: executives, board members and p...

French Supermarket Fridge Attack: What Our Data Shows and What It Means for You

Monadnock Cyber Intelligence and Analysis · 13 October 2026 On the night of Oct. 11–12, attackers hijacked refrigeration controls in hundreds of French supermarkets, heated display cases to 50°C (122°F) and destroyed the stock. The warning was public 52 days earlier. The threat extends beyond one manufacturer. For U.S. organizations, the greater risk is not exposed equipment but the contractors who service it: the same door attackers used against Target in 2013. This time they came to break things, not to steal. An Intermarché supermarket in Remollon, France, in 2014. Intermarché reported about 300 stores hit in the attack; this photo predates it and does not show an affected store. Editorial credit: Alastair Wallace / Shutterstock.com What happened About 300 Intermarché and Netto stores were hit, along with 80 Coopérative U, 80 Carrefour, about 30 Monoprix and "several" E.Leclerc stores (The Hacker News, BleepingComputer). The attackers turned the equipment against it...