Screenshot of the fake website
With the rapid growth of AI-driven platforms, cybercriminals have wasted no time in exploiting the popularity of DeepSeek, a powerful open-weight reasoning large language model (LLM). According to a recent report from Kaspersky Labs, multiple threat actors are distributing Trojanized installers disguised as legitimate DeepSeek clients, tricking unsuspecting users into downloading stealer malware and backdoors.
Researchers at Kaspersky uncovered several fake websites impersonating DeepSeek’s official site, designed to lure users into downloading malicious software. Notably, these sites used domain names mimicking DeepSeek model versions, such as:
- r1-deepseek[.]net
- v3-deepseek[.]com
The fraudulent websites omit the option to start a chat, instead presenting users with a “Get DeepSeek App” button, leading to a malicious ZIP archive download. “However, the real DeepSeek doesn’t have an official Windows client,” Kaspersky researchers noted.
Clicking the button downloads “deep-seek-installation.zip,” containing an LNK shortcut file that, when executed, triggers a remote script download. This script then retrieves and runs additional payloads, including:
- Python-based stealer malware
- A renamed svchost.exe (actually a modified python.exe) to evade detection
- Data exfiltration tools targeting cookies, login credentials, cryptocurrency wallets, and personal files
Once executed, the malware stealthily transmits stolen data via a Telegram bot or Gofile file-sharing service, making it difficult for victims to notice the breach.
Threat actors continuously adapt their tactics—Kaspersky notes that one of the malicious DeepSeek sites switched tactics mid-investigation. At one point, the v3-deepseek[.]com domain began distributing a fake Grok AI client, an AI model developed by xAI.
Additionally, attackers leveraged social media manipulation to amplify their reach. Kaspersky researchers discovered posts on X (formerly Twitter) promoting fake DeepSeek installers, some gaining over 1.2 million views. Most of the reposts were likely automated bots, researchers noted.
Beyond the stealer malware, Kaspersky observed targeted backdoor campaigns aimed at more technically advanced users. Attackers disguised malicious payloads as DeepSeek-compatible frameworks like Ollama, making them appear legitimate to AI developers.
Some fake DeepSeek domains, such as:
- app.delpaseek[.]com
- app.deapseek[.]com
delivered a custom backdoor detected as Backdoor.Win32.Xkcp.a. This malware establishes a KCP tunnel, granting attackers persistent remote access to compromised machines.
Other malware samples, disguised within deep_windows_Setup.zip, were observed using DLL sideloading and steganography to inject a Farfli backdoor variant, an advanced technique designed to evade security tools.
As AI technologies like DeepSeek gain traction, cybercriminals will continue to exploit the demand for AI tools by distributing fake installers and credential stealers. Kaspersky researchers urge users to verify software sources and avoid downloading AI clients from unofficial sites.
“Attempting to use the chatbot could result in the victim losing social media access, personal data, and even cryptocurrency,” the report warns.
Users and organizations should remain vigilant, implement endpoint security measures, and educate employees on AI-themed phishing tactics to prevent falling victim to these evolving cyber threats.
Related Posts:
- DeepSeek’s Exposed Database Leaks Sensitive User Information
- DeepSeek Disrupts AI Landscape: High Performance at Low Cost
- DeepSeek’s $6 Million AI Model Outperforms GPT-4
- Multiple Security and Privacy Flaws Found in DeepSeek iOS App
- Poseidon Stealer Malware Targets Mac Users via Fake DeepSeek Site
