Security & privacy

TypoPrompting: How fake AI domains hijack browsers and steal data

Researchers identified 186 lookalike AI domains distributing malware, search hijackers, and ClickFix lures under the campaign name TypoPrompting.

A keyboard with displaced keys leading to a trapdoor, symbolizing typosquatting risks.
Illustration generated for this article

Cybersecurity researchers at Island have uncovered a widespread campaign dubbed TypoPrompting, where attackers register misspelled domains mimicking popular artificial intelligence tools. These sites redirect users to malicious payloads, including browser hijackers and social engineering lures, exploiting the rapid adoption of AI services in daily workflows.

What happened

The campaign involves 186 fraudulent domains targeting 27 major AI brands, such as Anthropic Claude, OpenAI ChatGPT, Google Gemini, Microsoft Copilot, and DeepSeek. Some of these domains were registered as early as July 13, 2023, indicating a long-term strategy rather than a sudden spike. The attackers rely on typosquatting, registering addresses that differ by only one character from legitimate services, such as cloude[.]ai for Claude or chatlgpt[.]com for ChatGPT.

Island identified three distinct delivery mechanisms used by these domains. The first route employs ClickFix-style lures, where users land on fake Cloudflare verification pages that prompt them to copy and paste PowerShell commands or run executables. The second pathway directs visitors to fake "Security Check" pages that distribute a browser-hijacking malware known as Net Guard. The third route redirects traffic to a family of Chrome extensions that hijack search queries, collectively impacting nearly one million users.

This activity builds on existing abuse of parked domains. As noted by Infoblox in December 2025, domain owners can monetize empty sites by selling traffic to advertisers via traffic distribution systems. In this case, those systems are weaponized to profile devices and funnel users toward specific malware strains based on their operating system and browser fingerprint.

How it works

The ClickFix lure begins with a fingerprinting script that collects the user’s public IP address and browser details. Depending on the device, the site presents a tailored deception: Windows users see a fake verification page asking them to run a clipboard command, while macOS users are prompted to execute a shell command for a supposed security update. This method bypasses traditional download warnings by relying on user action to execute the payload directly in the terminal or command line.

The Net Guard malware operates differently by leveraging trust in official software repositories. The malicious version of Net Guard is byte-for-byte identical to the legitimate app available in the Microsoft Store, complete with a valid Microsoft signature. However, once installed, it activates hidden capabilities without user consent. It waits for the user to become idle, then modifies Chrome settings to change the default search engine and injects attacker-controlled JavaScript into browser sessions.

The search hijacking extension family, which includes Quick Search and GPT for Chrome, uses a simpler mechanism. These extensions do not require executable JavaScript or broad permissions to function. Instead, they modify the browser manifest to override the default search provider with a malicious endpoint like find.flashysearches[.]com. This redirects all search queries to a third-party server, potentially exposing sensitive internal data such as hostnames and project details.

Key details

  • The campaign targets 27 AI brands across 186 bogus domains, including lookalikes for Claude, Gemini, and Copilot.
  • Three primary attack vectors are used: ClickFix lures, Net Guard malware, and search-hijacking Chrome extensions.
  • Net Guard is a signed Microsoft Store app that secretly alters browser settings and runs remote JavaScript when the user is idle.
  • The search hijacker extensions have 972,000 declared users as of November 3, 2026, across 16 live Chrome Web Store listings.
  • Some domains in the network, such as cloude[.]ai, were registered in July 2023, showing long-term infrastructure preparation.
  • The ClickFix lures also target non-AI entities, spanning 211 entry domains including banks and enterprise software providers.

Why it matters

For software teams and IT leaders, this campaign highlights the risk of relying on manual URL entry for critical tools. As AI platforms become central to development and operations, the likelihood of typos increases, especially when developers are switching between multiple tabs and services. The use of signed, legitimate-looking applications like Net Guard demonstrates that traditional trust indicators, such as digital signatures, are no longer sufficient guarantees of safety.

The scale of the search hijacking operation poses a significant data leakage risk. With nearly one million users affected, many of whom may be in corporate environments, search queries containing proprietary code snippets, customer names, or internal infrastructure details are being exfiltrated. This passive data collection is difficult to detect because it does not involve obvious malware symptoms like system slowdowns or ransomware notes.

Furthermore, the integration of parked domain monetization with malware distribution complicates defense strategies. Traffic distribution systems dynamically route users based on real-time profiling, making static blocklists less effective. Security teams must now account for the possibility that a seemingly benign redirect from a mistyped URL could lead to a sophisticated, multi-stage infection chain tailored to their specific environment.

What you can do

  • Enforce the use of bookmarked links or internal portals for accessing AI tools to prevent typosquatting risks.
  • Audit installed Chrome extensions regularly and remove any that are not explicitly required for work tasks.
  • Monitor network traffic for unusual search provider redirects or connections to unknown domains like find.flashysearches[.]com.
  • Educate developers on the dangers of ClickFix lures and prohibit the execution of clipboard commands from unverified sources.
  • Implement DNS filtering to block known malicious domains and restrict access to unapproved traffic distribution systems.
  • Review browser policies to lock down default search engines and prevent unauthorized extensions from modifying core settings.
  • #cybersecurity
  • #ai security
  • #malware
  • #browser security
  • #typosquatting

Tools from the Bytechap store

Keep reading

All stories