Sogou Input Method Flaw Lets Hackers Install Backdoors

Security researchers at Gen Digital have revealed how a hacking group known as UNC3569 exploited a fundamental design weakness in Sogou Input Method. This tool is used by hundreds of millions of people to type Chinese characters on Windows. By sending a specially crafted link to victims, the attackers were able to install a persistent backdoor that gave them full control over the logged-in user's computer.

The Hacker News reported on this finding, which was published Thursday. The group, linked to China by Google Threat Intelligence, has targeted government, education, and finance sectors in East and Southeast Asia since 2021. While Tencent, the owner of Sogou, released a fix in April 2026, researchers warn that the underlying architectural issues remain largely unresolved in the current version.

The Attack Chain Relies on Unchecked Links

The vulnerability stems from how Sogou components communicate on Windows. The software registers a custom link type that allows external programs to trigger internal actions. When a user clicks a link, a helper program reads the command and passes along any arguments provided by the sender. Gen Digital found that this helper does not validate or filter these arguments, meaning an attacker can dictate exactly what the software should do.

In this specific attack, the malicious link instructed the Sogou settings program to open its internal skin store with a web address chosen by the attacker. The skin store is the only part of the application that opens a browser window. Because the code accepts any URL without verification, the attacker could force the browser to load a malicious page designed to exploit the underlying web engine.

Outdated Browser Engine Enables Full System Access

The critical failure lies in the browser engine embedded within Sogou. Gen Digital discovered that the software uses a version of Chromium dating back to March 2020. More importantly, two key security protections were explicitly disabled in the code. The sandbox, which isolates web pages from the rest of the system, was switched off. Additionally, the same-origin policy, which prevents a page from accessing data from other sites, was also disabled.

With these safeguards removed, a simple JavaScript flaw in the attacker's page translates directly into code execution on the user's computer. The attacker gains the same privileges as the logged-in user, allowing them to move files, install the GRAYRABBIT backdoor, and load additional malicious modules from their servers at will. There is no secondary barrier preventing the code from escaping the browser environment.

Tencent's Fix Blocks Entry but Leaves Core Risks

Tencent addressed the immediate threat by blocking the specific attack vector, but the root cause persists. In the patched version examined by Gen Digital, the built-in browser engine remains the 2020 version, and its sandbox is still disabled. This means that while the specific link-based exploit is blocked, the underlying architecture remains vulnerable to other methods of triggering the internal browser with untrusted content.

There is also a disagreement on how easily this attack can be executed. Gen Digital argues that clicking the link was sufficient to trigger the compromise. Tencent, however, describes the chain as complex, suggesting that an attacker would need to use social engineering to get users to manually authorize a browser pop-up prompt. Neither party has clarified what the victim actually sees during the interaction, leaving uncertainty about the level of user awareness required for the attack to succeed.