The rapid tightening of Chinese censorship demands has plunged local internet companies into a high-stakes game of survival, with executives knowing they face closure if they allow the distribution of banned content.
“It’s no exaggeration to say that any company could die overnight,” said Chen Taifeng, a vice-president at Yixia Technology, the developer of MiaoPai, a hugely popular service for live-streaming and video-sharing in the country.
It has become an industry joke that “almost all of the people in charge of content-audit departments have white hair.”
Those departments are charged with enforcing government censorship edicts, which have grown increasingly strict as China seeks to ensure internet video abides by standards more similar to those that have long governed broadcast television.
The risks of non-compliance are real: Last year, new regulations required live broadcasters to ensure content is “beneficial to the promotion of socialist core values.” Numerous broadcasters have since been inspected, and many shuttered. In April alone, regulators yanked 370 live-streaming apps from online stores.
To stay out of trouble, companies are pushing self-censorship beyond government demands, using people and sophisticated machine learning to scour videos and pictures for offences.
An internal document obtained by The Globe and Mail illuminated the sweep of government censorship demands, including bans on flirtatious dancing, “vulgar” use of microphones, Western critiques of Chinese political theory, the names or pictures of leaders and their families and the results of environmental investigations.
But Chinese companies are largely left to police such content on their own, and the increasingly stringent requirements have created an atmosphere of fear in a country where hundreds of millions of people now tune in to live online broadcasts every month.
The Globe and Mail recently attended a meeting during which executives from several live-broadcasting companies offered a surprisingly detailed look inside the Chinese censorship regime and how companies are enforcing rapidly shifting rules.
Yixia Technology has created a special department “to cope with the complicated rules and make sure we have the latest information,” Mr. Chen said. In addition to using automated screening software powered by artificial intelligence, the company hires human auditors who monitor screens displaying 30 live streams at a time.
Companies must also manage content that might have been considered acceptable only months ago, but which now violates new rules, said Zhou Tian, vice-president of security at Meitu, China’s biggest app for selfies.
Meitu has a photo-sharing service, and in examining older content, “we have found that some things that wouldn’t even have been called problems in the past were raised to a new level of seriousness,” he said.
Xiuse Entertainment, a live-broadcasting company, has designed software to spot abnormalities, such as a user whose audience suddenly grows or language that rapidly increases in popularity.
“The directors of all departments need to stay alert 24/7, in case anything bad happens,” said Zhou Gaoqin, director of research and development at Xiuse, who acknowledged that the company, to be safe, sets in place content restrictions stricter than those demanded by government.
In the past year, internet companies in China have also hired thousands of workers to police content. But that work takes a toll. “Content auditing is a very time-consuming and boring job,” Mr. Chen said. “It also places them under heavy mental pressure.”
The China Association of Performing Arts, for example, has set up black and grey lists. Individuals that “break the bottom line” by broadcasting pornography or terrorism will be blacklisted across the industry and permanently cut off. The grey list “is for those who occasionally break regulations without consequences that are too terrible,” said Qu Tao, an official with the association. They can be suspended from live broadcasts for three months.
Companies are also increasingly turning to digital nannies, which use China’s burgeoning sophistication in image recognition to keep up with shifting government demands. Among the leaders in China is Tuputech, a Shanghai-based company whose servers process more than a billion images a day for some of China’s most popular online services, including MiaoPai, video-sharing site bilibili and gay social app Blued.
For Tuputech, meanwhile, China’s censorship regime has created market demand to perfect image-recognition skills that chief executive Li Mingqiang says now exceed the capabilities of Western firms such as Facebook and Google when it comes to spotting banned content.
Tuputech has tuned its technology to recognize obscenity, pornography, violence, subculture symbols and politically sensitive content, such as Tibetan flags and images of political leaders, Mr. Li said in an interview.
One of the company’s boasts is that it can respond to new rules with great speed. “Even if you wanted to block [images of] elderly people dressed in blue and walking on the beach, we could do that,” he said.
Critics have raised alarm over the increasing reach of the Chinese government into the expression of its people, and the advent of a new form of digital authoritarianism that uses machines to expand radically the power authorities wield.
Mr. Li has played a central role in the building of China’s online space. He was one of the creators of WeChat, the chat and payment app that is so widely used that it effectively acts as the internet for many Chinese smartphone users. WeChat is widely believed to offer a transparent environment for Chinese authorities to monitor communications, but Mr. Li argues that purpose is positive.
“The Chinese government has good intentions in imposing these restrictions on content,” he said.
“They do it out of concern for people’s safety and social harmony. Their actions are beyond reproach.”
With reporting by Alexandra Li