According to people with firsthand knowledge of the activities, TikTok is cloning its recommendation algorithm for its 170 million U.S. users, which may make it more acceptable to American lawmakers who want to ban it
The source code separation efforts commissioned by ByteDance, the Chinese parent company of TikTok, in late 2017 preceded the emergence of a bill in Congress this year that sought to compel the sale of TikTok’s U.S. operations. In April, the measure was formally ratified into law.
According to anonymous sources, who were not authorized to discuss the short-form video-sharing application publicly, the division of the code may create the conditions for the sale of U.S. assets. However, no such intentions exist at this time.
As previously stated, the organization has no intention of divesting its U.S. assets, and such a course of action is implausible.
TikTok declined to comment initially.
“The Reuters story published today is misleading and factually inaccurate,” TikTok stated in a post on X that opens a separate tab after publication.
The post should have specified which aspects of the story were inaccurate.
TikTok additionally published a segment from its federal lawsuit:
“The ‘qualified divestiture’ that the Act requires to permit TikTok to continue operations in the United States is technologically, commercially, and legally unattainable.”
“Certainly not within the 270-day timeframe required by the legislation.”
“Our reporting stands firm,” stated a Reuters spokesperson.
In May, TikTok and its Chinese parent company, ByteDance, filed a federal lawsuit in the United States, attempting to obstruct the law mandating the app’s sale or prohibition by January 19.
On Tuesday, an appeals court in the United States expedited the schedule for hearing legal challenges to the new law.
MILLION-THOUSAND CODE LINES
Hundreds of ByteDance and TikTok engineers in the United States and China were tasked with segregating millions of lines of code containing the organization’s algorithm that matches users with videos that suit their preferences over the past few months.
According to two sources familiar with the project, the engineers aim to create a unique code base that operates independently of the systems used by Douyin, ByteDance’s Chinese version of TikTok. Additionally, they aim to remove any data that may associate the base with Chinese users.
The hitherto undisclosed strategy offers an exceptional insight into the potential technical segregation of TikTok’s operations in the United States. It demonstrates the extent to which TikTok is willing to mitigate the bipartisan political risk it encounters.
Proponents of the legislation, including U.S. President Biden, contend that TikTok grants Beijing excessive access to vast quantities of data, which could potentially be exploited to surveil or sway TikTok’s American users.
Before this, Reuters stated that selling the application containing algorithms is extremely unlikely. In 2020, content recommendation algorithms were included on the export-control list of the Chinese government.
Consequently, TikTok’s algorithm became subject to administrative licensing procedures, necessitating its sale or divestiture.
According to a legal filing, ByteDance engineers initially created the source code for TikTok’s recommendation engine in China.
They modified it to meet the operational requirements of TikTok’s various international markets, including the United States.
ByteDance has attributed TikTok’s prevalence to its recommendation engine’s efficacy, which generates personalized content feeds for users based on their engagement with the content they view.
“DOUBLE SOURCE”
The arduous nature of the task, which the sources labeled “dirty work” and described as complex, highlights the challenge associated with decoupling the underlying code that connects TikTok’s operations in the United States to its Chinese parent.
According to these sources, the work is anticipated to require more than a year to finish.
TikTok and ByteDance have promised to prosecute United States law based on the First Amendment. According to the sources, engineers continue to be directed to decouple the U.S. recommendation engine of TikTok from the more extensive network of ByteDance.
U.S. legislators and regulators rejected Project Texas, an earlier attempt to isolate user data from the country. Presently, the organization is endeavoring to intensify its endeavors to demonstrate that its operations in the United States are autonomous from those of its Chinese proprietor.
According to the sources, TikTok executives contemplated open-sourcing a portion of the platform’s algorithm or making it accessible and adaptable to others to demonstrate technological transparency.
The plans and updates regarding the code-splitting project were disseminated by executives through various means, including an internal communications system known as Lark, internal planning documents, and a team all-hands session, according to one source who attended the meeting and another who reviewed the messages.
According to one source, compliance and legal concerns regarding the transferability of code segments to TikTok are adding complexity to the endeavor. Before each line of code can be incorporated into the separate code base, it must undergo a review process, according to the sources.
The objective is to establish a fresh source code repository for a recommendation algorithm exclusive to TikTok U.S. TikTok U.S. will operate and sustain its recommendation algorithm independently of TikTok apps in other regions and its Chinese counterpart Douyin once the transition is complete.
According to sources, that action would deprive the company of its parent company’s enormous engineering development capabilities in Beijing.
According to additional sources, TikTok management is aware that TikTok U.S. might need to be more capable of providing the same degree of performance as the current TikTok if the separation of the recommendation engine from its Chinese counterpart is finalized.
TikTok U.S. heavily depends on ByteDance’s engineers in China to update and maintain the code base to optimize user engagement.