AI-Assisted Phishing Campaign Exploits Browser Permissions to Capture Victim Data

Executive Summary

Cyble Research & Intelligence Labs (CRIL) has identified a widespread, highly active social engineering campaign hosted primarily on edgeone.app infrastructure.

The initial access vectors are diverse — ranging from “ID Scanner,” and “Telegram ID Freezing,” to “Health Fund AI”—to trick users into granting browser-level hardware permissions such as camera and microphone access under the pretext of verification or service recovery.

Upon gaining permissions, the underlying JavaScript workflow attempts to capture live images, video recordings, microphone audio, device information, contact details, and approximate geographic location from affected devices. This data is subsequently transmitted to attacker-controlled infrastructure, enabling operators to obtain Personally Identifiable Information (PII) and contextually sensitive information. 

Further analysis revealed indicators of potential AI-assisted code generation, including structured annotations and emoji-based message formatting embedded within the operational logic. These characteristics reflect a growing trend where threat actors leverage generative AI tools to accelerate the development of phishing frameworks.

The breadth of data collected in this campaign extends beyond traditional credential phishing and raises significant security concerns. Harvested multimedia and device telemetry could be leveraged for identity theft, targeted social engineering, account compromise attempts, or extortion, posing risks to both individuals and organizations. (Figure 1)

Key Takeaways

• Infrastructure: Extensive use of edgeone.app (EdgeOne Pages) for hosting low-cost, scalable, and highly available phishing landing pages.
• Biometric Harvesting: The operation abuses legitimate browser APIs to access cameras, microphones, and device information after user consent.
• C2 Mechanism: Utilization of the Telegram Bot API (api.telegram.org) as a streamlined C2 and data exfiltration channel.
• Diverse Lures: Attackers rotate lures, including “ID Scanner” and “Health Fund AI”, to target various demographics and bypass regional security filters.
• The phishing pages impersonate popular platforms and services, including TikTok, Telegram, Instagram, Chrome/Google Drive, and game-themed lures such as Flappy Bird, to increase victim trust.
• Once interaction occurs, the campaign attempts to collect multiple forms of sensitive data, including photographs, video recordings, microphone audio, device information, contact details, and approximate geographic location.

Overview

• Campaign Start: Observed since early 2026
• Primary Objective: Harvesting victim multimedia data and device information
• Primary Infrastructure: edgeone.app (multiple subdomains)
• Impersonated Brands: TikTok, Telegram, Instagram, Chrome/Google Drive, Flappy Bird
• Key Behavior: Browser permission prompts used to capture camera images, record audio/video, enumerate device metadata, retrieve geolocation information, and attempt contact list access through browser APIs.

The campaign operates as a web-based phishing framework that captures photographs directly from victims’ devices. The infrastructure hosts multiple phishing templates that impersonate verification systems or service recovery portals. The goal is to socially engineer users into granting browser permission for camera access.

Unlike traditional credential phishing pages, these pages do not primarily collect typed input. Instead, they rely on browser hardware permissions, requesting access to the device’s camera. Once permission is granted, the page silently captures a frame from the live video stream and exfiltrates it.

The use of Telegram as a data collection mechanism indicates that the operators prioritize low operational complexity and immediate access to stolen data. Since Telegram bots can receive file uploads through simple HTTP requests, attackers can directly integrate the API into client-side scripts.

Business Impact and Potential Abuse

The data collected through this campaign provides attackers with multiple forms of sensitive personal information and contextual intelligence, thereby significantly increasing the effectiveness of follow-on attacks.

One potential abuse scenario involves identity fraud and account recovery manipulation. The campaign captures victim photographs, video recordings, and audio samples that could be used to bypass identity verification workflows used by financial platforms, social media services, or other online services that rely on biometric or video-based verification.

Additionally, the collection of device information, location data, and contact details allows attackers to build detailed victim profiles. This information may be used to perform targeted social engineering attacks, impersonate victims in communication platforms, or craft convincing fraud attempts against their contacts.

Another concerning use case involves extortion and intimidation. Because the campaign captures multimedia data, such as camera images, video recordings, and microphone audio, attackers may pressure victims by threatening to expose the collected material unless a payment is made.

For organizations, the broader business impact includes:

• Increased risk of identity theft and account takeover attempts
• Potential abuse of stolen biometric and multimedia data in fraud schemes
• Targeted phishing or fraud campaigns against employees and customers
• Reputational damage if impersonated brand identities are used in malicious campaigns

The campaign’s ability to collect multiple categories of sensitive information from a single interaction significantly amplifies the risk to both individuals and businesses.

Why does this matter?

This campaign marks a significant evolution in phishing operations, shifting from credential theft to harvesting biometric and device-level data. By abusing browser permissions to capture victims’ live images, audio, and contextual device information, threat actors can obtain high-quality identity data that is difficult to revoke or replace.

The stolen data can be leveraged to bypass video-KYC and remote identity verification processes, enabling fraudulent account creation, synthetic identity fraud, account takeover, and financial scams across banking, fintech, telecom, and digital service platforms. Additionally, high-resolution facial images and audio samples may be weaponized for AI-driven impersonation and deepfake attacks, increasing the effectiveness of business email compromise and targeted social engineering campaigns.

For organizations, the campaign introduces elevated risks, including financial losses, regulatory non-compliance, AML exposure, reputational damage, and erosion of trust in digital onboarding systems, highlighting the growing need for stronger verification controls and browser-permission abuse detection.

Technical Analysis

The infection chain, as outlined in Figure 2, shows the stages of the attack.

Phishing Page Behaviour

The phishing page contains embedded JavaScript that leverages browser media APIs to access the victim’s device camera after obtaining user permission. Once access is granted, the script initializes a live video stream and processes its frames.

A capture function then renders a frame from the video feed onto an HTML5 canvas using ctx.drawImage(), effectively converting the live camera input into a static image. (see Figure 3)

The canvas content is subsequently encoded into a JPEG blob via canvas.toBlob(), creating a binary image object that can be transmitted through HTTP requests to attacker-controlled infrastructure.

Expanded Data Collection Capabilities

Analysis of the campaign script indicates that the phishing framework performs extensive device fingerprinting and environment enumeration before initiating camera-based verification workflows.

The script collects system metadata using the following browser APIs

• navigator.userAgent
• navigator.platform
• navigator.deviceMemory
• navigator.hardwareConcurrency
• navigator.connection
• navigator.getBattery

This allows the attacker to gather detailed information such as operating system type and version, device model indicators, screen resolution and orientation, browser version, available RAM, CPU core count, network type, battery level, and language settings.

Additionally, the script retrieves the victim’s public IP address using services such as api.ipify.org, then enriches the geolocation using ipapi.co, enabling the collection of country, city, latitude, and longitude data. (see Figure 4)

This telemetry is aggregated and transmitted to the attacker via the Telegram Bot API, providing operators with contextual information about the victim’s device and location prior to further data harvesting.

Beyond system profiling, the script implements multiple routines for collecting multimedia and personal data via browser permission prompts. The campaign captures several still images from both the front-facing and rear-facing cameras, records short video clips using the MediaRecorder API, and performs microphone recordings.

These recordings are packaged as JPEG, WebM video, or WebM audio files and exfiltrated via Telegram API methods such as sendPhoto, sendVideo, and sendAudio. (see Figure 5)

Additionally, the script attempts to access the victim’s contact list through the Contacts Picker API (navigator.contacts.select), requesting attributes such as contact names, phone numbers, and email addresses. If granted, the selected contacts are formatted into structured messages and transmitted to the attacker. (see Figure 6)

User Interface Manipulation

The phishing pages include interface elements designed to convince victims that the image capture process is legitimate.

For example, status messages displayed during execution may include:

• “Capturing photo”
• “Sending to server”
• “Photo sent successfully”

These messages simulate the behavior of legitimate identity verification platforms and help maintain the illusion that the process is part of a valid verification workflow.

Once the image is successfully transmitted, the script terminates the camera stream and resets the interface after a short delay.

Infrastructure Observations

Analysis of the campaign revealed that the phishing pages are primarily hosted under the edgeone.app domain. Multiple variations of phishing pages were observed using similar JavaScript logic and workflow patterns.

The consistent use of the same infrastructure suggests that attackers may be operating a templated phishing kit capable of generating different themed pages while maintaining the same underlying data-collection logic.

Because the image exfiltration occurs through Telegram infrastructure, the phishing pages themselves do not require backend servers, simplifying deployment and enabling rapid rotation of phishing URLs.

Indicators of Potential Generative AI Use in Script Development

During analysis of the phishing framework, researchers observed the use of emojis embedded directly within the script’s message formatting logic. These emojis appear in structured status messages that are assembled and transmitted during the data collection workflow. The use of decorative Unicode symbols within operational code is uncommon in manually written malicious scripts but has increasingly been observed in campaigns that use generative AI tools during development. (see Figure 7)

Targeted Countries and Impersonated Brands

During infrastructure monitoring and phishing URL telemetry analysis, the campaign’s infrastructure appears to be globally accessible. Analysis of the phishing templates used in this campaign reveals that the operators impersonate a range of widely recognized consumer platforms and applications. Observed brand impersonation themes include:

Impersonated Brand	Observed Theme
TikTok	Free followers/engagement rewards
Flappy Bird	Game reward or verification workflows
Telegram	Account freezing or verification alerts
Instagram	Account recovery or follower reward systems
Google Chrome / Google Drive	Security verification prompts

Conclusion

Our deep-dive analysis revealed a sophisticated phishing campaign that extends beyond traditional credential theft by harvesting multimedia and device-level data through browser permission abuse.

The campaign attempts to collect photographs, video recordings, audio recordings from microphones, contact details, device information, and approximate location data directly from victims. This operation demonstrates a growing trend where attackers leverage client-side scripting and legitimate web services to collect and transmit sensitive data without relying on traditional command-and-control infrastructure.

Indicators in the script also suggest AI-assisted development, reflecting how threat actors may be using generative AI tools to accelerate the creation of phishing frameworks.

The breadth of information collected increases the potential for identity theft, targeted social engineering, account compromise attempts, and extortion. Organizations should remain cautious about phishing pages that request hardware permissions, such as camera, microphone, or contact access, particularly when originating from untrusted domains.

Cyble’s Threat Intelligence Platforms continuously monitor emerging threats, attacker infrastructure, and malware activity across the dark web, deep web, and open sources. This proactive intelligence empowers organizations with early detection, brand and domain protection, infrastructure mapping, and attribution insights. Altogether, these capabilities provide a critical head start in mitigating and responding to evolving cyber threats.

Our Recommendations

We have listed some essential cybersecurity best practices that serve as the first line of defense against attackers. We recommend that our readers follow the best practices given below:

• Restrict camera permissions for unknown websites
• Monitor outbound traffic to api.telegram.org when originating from browser sessions
• Deploy browser security extensions capable of identifying phishing pages
• Implement domain monitoring for suspicious infrastructure hosting phishing kits

MITRE ATT&CK® Techniques

Tactic	Technique ID	Procedure
Initial Access	T1566 – Phishing	Phishing pages used to lure victims to malicious verification workflows.
Execution	T1059.007 – JavaScript	Malicious JavaScript executed in the victim’s browser.
Collection	T1125 – Video Capture	Camera access is used to capture photos and videos of victims.
Collection	T1123 – Audio Capture	Microphone access is used to record the victim’s audio.
Collection	T1005 – Data from Local System	Device information is collected from the browser environment.
Collection	T1213 – Data from Information Repositories	Contact details retrieved from the device contact list.
Discovery	T1082 – System Information Discovery	Device and browser information enumeration.
Discovery	T1614 – System Location Discovery	Victim IP and geographic location collected.
Exfiltration	T1567 – Exfiltration Over Web Services	Collected data transmitted to the attacker’s infrastructure.

Indicators of Compromise (IOCs)

The IOCs have been added to this GitHub repository. Please review and integrate them into your Threat Intelligence feed to enhance protection and improve your overall security posture.

The post AI-Assisted Phishing Campaign Exploits Browser Permissions to Capture Victim Data appeared first on Cyble.

Cyble – ​Read More