A delivery pattern termed TypoPrompting exploits ordinary typing errors when users attempt to access popular artificial intelligence services. The campaign targets lookalike domains for tools such as Claude, ChatGPT, and DeepSeek, as well as creative platforms like Midjourney and ElevenLabs. These domains are registered to capture traffic from users who make minor spelling mistakes, such as swapping adjacent keys or omitting characters. The infrastructure spans hundreds of entry points, including lookalikes for financial institutions and enterprise software. The risk involves redirecting legitimate professional workflows into malicious environments where visitors are prompted to execute commands or install unauthorized software, effectively turning a simple navigation error into a security incident.
The mechanism begins with traffic distribution systems that profile visitors based on location, operating system, and browser type to select compatible lures. For Windows users, the process leads to a fake verification page that instructs the victim to paste a specific command into the Run dialog. This command downloads a hidden PowerShell script from a decimal-encoded IP address, bypassing standard pattern filters, and executes it with execution policy bypassed. A separate route delivers a Microsoft Store-signed application that appears legitimate but contains a browser-hijacking engine. This application uses remote configuration to change default search engines via synthetic input, execute remotely supplied scripts through developer tools, and modify browser profiles. The infrastructure also includes search-hijacking extensions that override default search providers without requiring declared permissions, routing all address-bar queries through third-party endpoints to capture sensitive internal data.[/subscribe_to_unlock_form]
A delivery pattern termed TypoPrompting exploits ordinary typing errors when users attempt to access popular artificial intelligence services. The campaign targets lookalike domains for tools such as Claude, ChatGPT, and DeepSeek, as well as creative platforms like Midjourney and ElevenLabs. These domains are registered to capture traffic from users who make minor spelling mistakes, such as swapping adjacent keys or omitting characters. The infrastructure spans hundreds of entry points, including lookalikes for financial institutions and enterprise software. The risk involves redirecting legitimate professional workflows into malicious environments where visitors are prompted to execute commands or install unauthorized software, effectively turning a simple navigation error into a security incident.
The mechanism begins with traffic distribution systems that profile visitors based on location, operating system, and browser type to select compatible lures. For Windows users, the process leads to a fake verification page that instructs the victim to paste a specific command into the Run dialog. This command downloads a hidden PowerShell script from a decimal-encoded IP address, bypassing standard pattern filters, and executes it with execution policy bypassed. A separate route delivers a Microsoft Store-signed application that appears legitimate but contains a browser-hijacking engine. This application uses remote configuration to change default search engines via synthetic input, execute remotely supplied scripts through developer tools, and modify browser profiles. The infrastructure also includes search-hijacking extensions that override default search providers without requiring declared permissions, routing all address-bar queries through third-party endpoints to capture sensitive internal data.[emaillocker id="1283"]
The campaign demonstrates active development and expansion, with infrastructure records dating back to 2023 and recent additions of new lookalike domains and redirectors. The scale includes over one hundred and eighty-six lookalike domains across twenty-seven brands, with a significant portion of the search-hijacking extensions still listed and displaying nearly one million declared users. The motivation appears to be both financial, through affiliate referrals and ad networks, and malicious, through the distribution of stealer malware and browser control tools. Defensive implications require organizations to monitor for browser-to-shell handoffs, verify installer provenance beyond store signatures, and centrally manage search providers and extension policies to prevent unauthorized modifications to user environments.
| Tactic | Technique Id | Technique | Sub-technique |
|---|---|---|---|
| Initial Access | T1566.002 | Phishing | Spearphishing Link |
| Initial Access | T1566.004 | Phishing | Spearphishing Voice |
| Initial Access | T1566.003 | Phishing | Spearphishing via Service |
| Execution | T1059.001 | Command and Scripting Interpreter | PowerShell |
| Persistence | T1547.001 | Boot or Logon Autostart Execution | Registry Run Keys / Startup Folder |
| Defense Evasion | T1027.001 | Obfuscated Files or Information | Encoding/Decoding |
| Defense Evasion | T1562.001 | Impair Defenses | Disable or Modify Tools |
| Credential Access | T1555.001 | Credentials in Files | Credentials in Web Browser Files |
| Discovery | T1082 | System Information Discovery | - |
| Collection | T1056.004 | Input Capture | Keylogging |
| Command and Control | T1071.001 | Application Layer Protocol | Web Protocols |
| Command and Control | T1071.004 | Application Layer Protocol | File Transfer Protocols |
| Objective | Behavior ID | Behavior |
|---|---|---|
| Anti-Behavioral Analysis | B0001 | Debugger Detection |
| Command and Control | B0030 | C2 Communication |
| Execution | E1059 | Command and Scripting Interpreter |
| Defense Evasion | E1112 | Modify Registry |
| Impact | E1510 | Clipboard Modification |
| Communication Micro-objective | C0002 | HTTP Communication |
| Communication Micro-objective | C0011 | DNS Communication |
| Data Micro-objective | C0026 | Encode Data |
| Operating System Micro-objective | C0036 | Registry |
The following reports contain further technical details:
[/emaillocker]