Description
Phishing campaigns traditionally rely on HTML text, links, input fields and forms to imitate legitimate services. Security products can inspect these elements to identify suspicious content, credential-harvesting forms and malicious redirects.
Unit 42's November 4, 2026 report describes three campaigns that reduce the visibility of phishing content by rendering it as graphics rather than ordinary HTML text.
The campaigns use three different approaches:
Delivering a complete phishing verification card as a remotely hosted image.
Drawing fake financial documents and receipts on HTML5 canvas elements.
Constructing QR codes through SVG vector definitions and coordinate-based rendering.
The common technique is to make the visual content presented to victims different from what conventional DOM inspection reveals. A webpage may appear to contain a detailed financial receipt or login prompt while exposing few or no corresponding text elements in its HTML structure.
Technical Description
The Document Object Model (DOM) represents the structure and content of a webpage. Security inspection tools often examine DOM elements, visible text, form fields, links and scripts to identify phishing behavior.
The campaigns described by Unit 42 manipulate this relationship between webpage appearance and underlying structure.
Graphics-based phishing refers to presenting deceptive content through graphical elements instead of conventional HTML text and form structures.
The three observed implementation methods are:
| Technique | Implementation | Intended effect |
|---|---|---|
| Uploaded image | Phishing content delivered as a preview image | Hides lure text from ordinary DOM inspection |
| HTML5 canvas | JavaScript draws fake documents and receipts | Displays graphical content without equivalent HTML text |
| SVG vector paths | QR codes assembled from reusable vector shapes | Conceals QR-code construction inside SVG definitions and coordinate references |
These techniques do not make phishing pages inherently undetectable. Instead, they can reduce the effectiveness of detection approaches that depend heavily on inspecting readable HTML content.
Attack Overview
Unit 42 tracked three campaigns from May through October 2026, recording a combined 1,740 detections.
Campaign 1: Single-Image Preview Lures
<text>Detection count: 1,188</text>
This campaign abuses an online collaboration and drawing-preview platform to display phishing content as a single image.
Victim experience
The victim receives or opens a drawing-preview link.
The preview displays an online retailer's account-verification card.
The card contains a yellow Continue Verification button.
A randomized document title is used to make the page appear legitimate.
The victim is encouraged to interact with the verification prompt.
Technical implementation
Unit 42 examined an 83,214-byte top-frame DOM containing:
Zero visible words.
Zero input tags.
Zero canvas tags.
One empty SVG wrapper of approximately 1.4 KB.
Instead of embedding the verification text and interface directly into the webpage, the page retrieves the complete verification card as a 1200 × 630 preview image from a cloud-services-provider URL.
The image is then displayed inside the preview container.
This approach means that the phishing message is visually available to the victim but is not represented as ordinary readable text in the page's HTML markup.
Monthly detections
| Month | Detections |
|---|---|
| May 2026 | 178 |
| June 2026 | 231 |
| July 2026 | 274 |
| August 2026 | 153 |
| September 2026 | 196 |
| October 2026 | 156 |
| Total | 1,188 |
July recorded the highest number of detections in this campaign.
Campaign 2: HTML5 Canvas-Painted PDF and Receipt Lures
<text>Detection count: 361</text>
The second campaign uses a fake PDF viewer to display blurred bank-transfer receipts and financial documents.
The pages imitate document-viewing experiences, encouraging victims to interact with what appears to be an encrypted or protected PDF.
Technical implementation
Unit 42 analyzed an 18,666-byte HTML document with the following characteristics:
Sets
robotsmetadata tonoindex, nofollow, noarchive.Uses
no-referreras the referrer policy.Loads custom PDF.js stylesheets.
Dynamically loads JavaScript ES modules named
download-dialog.mjsandapp.mjs.Uses
document.createElement('script')to inject module scripts.
The fake PDF viewer is structured inside:
div#viewer-container > div.document-preview-frame > div#pdf-viewer.pdfViewer
The page renders a blurred interbank wire-transfer receipt onto a 974 × 1260 HTML5 canvas, containing 1,227,240 square pixels.
A second hidden canvas element is also present.
An overlay button is displayed above the rendered document. Unit 42 identified the following buttons:
button.preview-eye[data-open-download]button#open-file
Credential-harvesting mechanism
When the victim clicks the preview or open-file button, the page invokes the native HTML5 showModal() function to display a password dialog.
The dialog contains:
form#password-forminput#pdf-password[type="password"][autocomplete="off"][required]
The victim is asked to enter a password under the pretext of unlocking an encrypted PDF.
The campaign also includes a separate app-download dialog containing an a#download-app link, which can be used to deliver a mobile payload.
This combines document-themed social engineering, password collection and a potential mobile-payload delivery mechanism within a fake PDF-viewing experience.
Monthly detections
| Month | Detections |
|---|---|
| May 2026 | 42 |
| June 2026 | 37 |
| July 2026 | 56 |
| August 2026 | 52 |
| September 2026 | 86 |
| October 2026 | 88 |
| Total | 361 |
October recorded the highest monthly detection count for this campaign.
Campaign 3: SVG Coordinate-Grid QR-Code Lures
<text>Detection count: 191</text>
The third campaign uses SVG vector graphics to construct QR codes that imitate the MAX Messenger web-login experience.
Instead of embedding a conventional QR-code image or drawing it on a canvas, the webpage constructs the QR code using SVG definitions, vector paths and coordinate transformations.
Technical implementation
Unit 42 analyzed a 322,322-byte DOM associated with a SvelteKit application.
The page imports JavaScript resources including:
/_app/immutable/entry/start.CP8bQY5P.js/_app/immutable/entry/app.BHnh9mMq.js
It also registers /service-worker.js through a Trusted Types policy named sveltekit-trusted-types.
The webpage impersonates the MAX Messenger web-login interface and presents a QR-code sign-in experience.
The examined DOM contains:
Zero image tags.
Zero canvas tags.
Zero input tags.
Only 27 visible words.
Six SVG elements totaling approximately 315,960 bytes.
Two notable SVG components include:
A hidden icon sprite,
svg#max-icons-sprite, measuring 251,859 bytes.A QR-code SVG measuring 62,532 bytes.
The QR-code SVG uses a viewBox of 0 0 4059 4059 and is displayed at 220 × 220 pixels.
QR-code construction
The SVG defines 11 modular vector shapes inside its defs section, including:
rect#rect-0path#empty-0path#b-0path#r-0path#l-0path#t-0path#lt-0path#lb-0path#rb-0path#rt-0Finder pattern
#point-0
The QR matrix is assembled through hundreds of coordinate-based transformations using:
g[transform="translate(X,Y)"] > use[xlink:href="#..."]
These elements are distributed across the 4059 × 4059 coordinate grid. The construction also overlays a center logo through:
clip-path="url(#logo-mask-0)"
This technique allows the QR code to be generated from reusable SVG components instead of a conventional image file.
The resulting visual experience imitates a legitimate MAX Messenger QR login page. QR-code login lures can expose victims to session-hijacking risks if they scan and authorize an attacker-controlled authentication flow.
Monthly detections
| Month | Detections |
|---|---|
| May 2026 | 20 |
| June 2026 | 54 |
| July 2026 | 34 |
| August 2026 | 22 |
| September 2026 | 29 |
| October 2026 | 32 |
| Total | 191 |
June recorded the highest monthly detection count for this campaign.
Comparative Campaign Analysis
| Campaign | Graphics technique | Primary lure | Detections |
|---|---|---|---|
| Campaign 1 | Hosted preview image | Retailer account verification | 1,188 |
| Campaign 2 | HTML5 canvas | Fake PDF and bank-transfer receipt | 361 |
| Campaign 3 | SVG vector paths | MAX Messenger QR-code login | 191 |
| Total | 1,740 |
Campaign 1 accounted for the largest share of detections. Campaign 2 focused on document and financial themes, while Campaign 3 used QR-code-based login deception.
The three campaigns demonstrate different ways of separating the visible phishing interface from conventional webpage text and form structures.
Indicators of Compromise (IOCs)
The following URLs were listed by Unit 42 as campaign indicators. They are defanged for safe handling.
Campaign 1 — Single-Image Preview Lures
| No. | Phishing URL |
|---|---|
| 1 | hxxps[:]//docs.google[.]com/drawings/d/1PkpYjKXZjijUIdTtGzC3gKAGrWM3SjTq89FJ6OZ9ULg/preview?JggAl |
| 2 | hxxps[:]//docs.google[.]com/drawings/d/1c8Beah-y6ykV4egbSfLqakuNr7rjk-etkS0Ee7ksdXY/preview?ZIB95 |
| 3 | hxxps[:]//docs.google[.]com/drawings/d/1pUWFVJkJJpoYqky3hzHMEMjb0bkAYM0NlgvxZYj4wik/preview?BcHpH |
| 4 | hxxps[:]//docs.google[.]com/drawings/d/1Lk8u8kvrgxHgjfoHl1NWcKp8sad3fCy4KcXzwLTEMyU/preview?uqrMX |
| 5 | hxxps[:]//docs.google[.]com/drawings/d/1d-rGOT7crc2txZ7vE_9xgsXXZNnIhbcaVbeH20YFNbI/preview |
| 6 | hxxps[:]//docs.google[.]com/drawings/d/1ufGqPEuWZK1uV__uSDP4gK84vgXJH3BTYw6WjIlbou0/preview?TNfEP |
| 7 | hxxps[:]//docs.google[.]com/drawings/d/1PjyZt0ZxiDFRo-FcrG5yB6bxeCyGfqHKqiIyUKidABI/preview?gbu6F |
| 8 | hxxps[:]//docs.google[.]com/drawings/d/1TrpKcBwrCxukLb1ajrMpkVhXjrr8N2exw-8Mv_AsYSw/preview |
Campaign 2 — Canvas-Based Document and Receipt Lures
| No. | Phishing URL |
|---|---|
| 1 | hxxps[:]//visualizadordepdf[.]com |
| 2 | hxxps[:]//faq-log-trezor-io.pages[.]dev |
| 3 | hxxps[:]//io-trezorstrtt.pages[.]dev/index.php?route=checkout/checkout |
| 4 | hxxps[:]//jjenimi.pages[.]dev/index.php?route=checkout/checkout |
| 5 | hxxps[:]//telegram-6q70dzzf.pages[.]dev |
| 6 | hxxps[:]//htxx-lgon.pages[.]dev |
| 7 | hxxps[:]//pub-434abe86553e4c99b30d1c3be588c762.r2[.]dev/f1nanc1al-sh33tsv2.3.html?centre= |
| 8 | hxxps[:]//profound-bubblegum-158420.netlify[.]app/ingreso.html |
| 9 | hxxps[:]//proviweb-d8764.firebaseapp[.]com |
| 10 | hxxps[:]//www.ky8032162[.]cc/index.html?shareName=www.ky804234[.]cc |
Campaign 3 — SVG QR-Code Login Lures
| No. | Phishing URL |
|---|---|
| 1 | hxxps[:]//perehod[.]website |
| 2 | hxxps[:]//verdepy-online.firebaseapp[.]com/cbmob/personas/index.html |
| 3 | hxxps[:]//moltravi-zekun39471628.vercel[.]app |
| 4 | hxxps[:]//submid-cotact136336.vercel[.]app |
| 5 | hxxps[:]//spark-login.vercel[.]app |
| 6 | hxxps[:]//solflare-mnemonic-to-keystore.pages[.]dev |
| 7 | hxxps[:]//comcast-account.gitbook[.]io/comcast-account-docs |
| 8 | hxxps[:]//gmininilogenix.gitbook[.]io/us/ |
| 9 | hxxps[:]//pub-fc87556782e141d88f0d4c70cc8e88ef.r2[.]dev/OfficialDocumentHandling.html |
IOC handling note: These are phishing URLs reported by Unit 42. The source does not provide file hashes, dedicated malware samples, or confirmed IP addresses for these campaigns. Hosting-platform domains appearing in the indicators should not automatically be treated as malicious in every context; investigate the specific URL and associated activity.
Potential Impact
The campaigns present several security risks:
1. Credential theft
The fake PDF campaign displays a password prompt designed to collect credentials under the pretext of opening an encrypted document.
2. Account compromise
Retailer verification lures can trick victims into submitting account credentials or sensitive authentication information.
3. Session hijacking
QR-code login lures can expose users to session-hijacking risks when victims scan and authorize a malicious login request.
4. Mobile payload delivery
The fake PDF campaign includes an app-download dialog that can deliver a mobile payload. The supplied Unit 42 report does not establish the exact payload's behavior or impact.
5. Evasion of conventional inspection
Because the phishing content is rendered graphically, detection systems relying primarily on visible HTML text and conventional form inspection may fail to identify important parts of the lure.
Recommendations
Immediate Actions
Block confirmed malicious URLs from the IOC list through appropriate web-filtering and security controls.
Review proxy, DNS and secure web gateway logs for connections to the reported phishing URLs.
Investigate suspicious login attempts following interactions with image-based verification pages, fake PDF viewers or QR-code login prompts.
Reset credentials for users confirmed to have submitted them to phishing pages.
Revoke active sessions and refresh authentication tokens where account compromise or session theft is suspected.
Investigate mobile application downloads initiated from suspicious document-viewing pages.
Preserve browser, proxy, endpoint and authentication logs for incident analysis.
Preventive Actions
Inspect rendered content: Extend phishing detection beyond DOM text extraction to include screenshots, image analysis and visual-content inspection.
Analyze canvas elements: Monitor pages that dynamically render financial documents, receipts or login interfaces using HTML5 canvas.
Inspect SVG structures: Examine SVG definitions, vector paths, embedded references and coordinate-based rendering for suspicious QR-code construction.
Strengthen QR-code security: Train users to verify the destination and authentication context before scanning QR codes or approving login requests.
Restrict suspicious downloads: Apply controls to mobile application downloads initiated from untrusted or unexpected document-viewing pages.
Improve browser protection: Use browser isolation, URL reputation services and anti-phishing controls to reduce exposure to suspicious preview links.
Enable phishing-resistant MFA: Prefer FIDO2 security keys or passkeys where supported, particularly for accounts with sensitive access.
Conduct user awareness training: Explain that image-based verification cards, unexpected encrypted-PDF prompts and QR-code login requests can be phishing lures.
Detection and Threat-Hunting Guidance
Security teams should avoid relying exclusively on HTML text and input-field inspection. The following behaviors can support investigation of similar campaigns.
| Area | Suspicious behavior to investigate |
|---|---|
| Browser activity | Unexpected drawing-preview links or document-viewer pages |
| Image rendering | Remote image retrieval used as the primary account-verification interface |
| Canvas activity | Canvas-rendered bank receipts or financial documents |
| Password dialogs | Password prompts displayed after clicking a document-preview button |
| Download behavior | Unexpected mobile application downloads from fake PDF viewers |
| SVG inspection | Large SVG structures containing QR-code paths and coordinate-based references |
| Authentication | Unusual login activity following QR-code scanning |
| Network activity | Connections to the phishing URLs listed in the IOC section |
Suggested investigation workflow
Retrieve the original URL and preserve the webpage and associated resources.
Capture the rendered page or browser screenshot for visual analysis.
Inspect the DOM for images, canvas elements, SVG definitions and dynamically loaded JavaScript.
Examine network requests to identify remotely hosted preview images and scripts.
Analyze QR-code destinations in an isolated environment without authorizing authentication requests.
Correlate browser activity with authentication logs to identify possible credential or session compromise.
MITRE ATT&CK Mapping
The following mappings are relevant analytical associations based on the behaviors described in the report. They are not presented as mappings explicitly assigned by Unit 42.
| Technique | ATT&CK ID | Relevance |
|---|---|---|
| Phishing: Spearphishing Link | T1566.002 | Delivery of malicious drawing-preview, document and QR-login URLs |
| User Execution: Malicious Link | T1204.001 | Victims are encouraged to open and interact with phishing links |
| Steal Web Session Cookie | T1539 | Relevant to potential session compromise through QR-code login lures |
| Input Capture: Web Portal Capture | T1056.003 | Relevant to credential collection through fake password prompts |
These mappings describe potential technique alignment. The report does not establish every downstream compromise behavior for all three campaigns.
Conclusion
The three phishing campaigns tracked by Unit 42 demonstrate how attackers are adapting the presentation layer of malicious webpages to evade traditional inspection techniques.
By using hosted images, HTML5 canvas rendering and SVG vector paths, attackers can display convincing phishing interfaces while keeping much of the visible content outside ordinary HTML text and form structures.
The 1,740 detections recorded between May and October 2026 highlight the importance of examining not only webpage source code but also how content is rendered and how users interact with it.
For defenders, effective phishing detection must combine DOM inspection with visual analysis, browser telemetry, network monitoring and authentication-event correlation. As phishing techniques become more graphics-driven, security controls must evolve beyond text-based detection.
