TikTok removed nearly 105 million videos for nudity and other policy violations in the first half of 2020

• TikTok has released its latest global transparency report, saying it removed almost 105 million videos in the first half of 2020 for policy violations.
• These included videos depicting nudity, or that endangered minors, among other policy breaches.
• Business Insider also learned that TikTok conducted an internal investigation following the spread of a video showing a man's death by suicide.
• TikTok says it wants to work with other Big Tech firms on a 'hashbank' of graphic content to prevent it spreading across platforms.

TikTok removed nearly 105 million videos for violating its rules in the first half of 2020, the company's latest transparency report reveals.

The number is double what the app removed in the final six months of 2019, but still accounts for less than one percent of the videos posted on the app — indicating the growth of the platform in the last year worldwide.

Like YouTube, which recently admitted its computer-driven video moderation system was producing too many false positives, TikTok "relied more heavily on technology" to monitor, flag and remove videos during the coronavirus pandemic.

Nearly 97% of the videos were removed proactively by TikTok, with 90.32% of them removed before receiving any views. Three in 10 of the offending videos were removed for containing adult nudity or sexual activities — a higher proportion than the previous six months — while a further 22.3% endangered the safety of minors.

Not included in the latest transparency report was the video of a 33-year-old man, Ronnie McNutt, dying by suicide, which spread across the app earlier this month. TikTok users rushed to warn each other about the content of the video to try and prevent its spread.

Business Insider has exclusively learned the outcome of an internal investigation into how the video went viral on the platform.

The video was first posted to the app shortly after McNutt's death was livestreamed on Facebook at the end of August, with a small number of videos receiving a limited number of views.

On the night of September 6, TikTok saw a massive spike in the number of videos showing McNutt's death posted on the app — the result of an organized campaign believed to be planned on the dark web. The attackers are also believed to have launched similar attacks on other platforms.

"This is an industry-wide challenge, which is why we have proposed to peers across the industry that we work together on creating a 'hashbank' for such violent, graphic content and warn each other when such content is discovered so that we can all better protect our users, no matter the app they use," said Theo Bertram, TikTok's director of government relations and public policy for Europe, who prefaced his comments by offering his deepest sympathies to the family and friends of McNutt.

To respond to the issue, TikTok's US interim chief executive Vanessa Pappas,today sent a letter to the heads of nine tech giants proposing a memorandum of understanding "that will allow us to quickly notify one another of such content."

Business Insider understands more than 10,000 versions of the video were posted to TikTok in a short period of time — far fewer than the 1.5 million versions of a video depicting the mass murder of Muslims at a mosque in Christchurch, New Zealand, that was spread widely on Facebook.

TikTok banned the accounts that posted versions of the video, many of which signed up specifically to upload it.

The idea that this was a "shallow but broad" attack, where many videos with a low number of views are posted to try and gain maximum exposure without tipping off the app's monitoring system, chimes with the thinking of a source with knowledge of ByteDance's moderation policies. They spoke in the immediate aftermath of the attack.

"If multiple accounts upload different edited versions of the video, it's hard to deal with," the person explained in early September. "A large number of videos with a small number of views accumulate a larger number of views."

Those trying to repost the video used a number of tools, including TikTok's filters and splicing tools, to bypass the app's algorithmic detection of the video.

Business Insider understands from the same person that videos are added to a deduplication library, which cross-checks videos posted to the app, removing matches.

At the same time, Business Insider understands many of the videos were viewed after users actively sought out the footage through the app's search tool and by clicking on specific profiles, rather than being served the video through TikTok's For You page.

This was an indication that some users actively wanted to find the footage after hearing about it, or that those wanting to disseminate the video knew that engagement with the video was more likely to surface the app to a wider number of users.

"Following an internal review, we found evidence of a coordinated effort by bad actors to spread this video across the internet and platforms, including TikTok," said Bertram.

"We detected and removed these clips for violating our policies against content that displays, praises, glorifies, or promotes suicide. We also took swift action including by banning accounts that were uploading this content."

In response to the spread of the suicide video, Business Insider understands TikTok has changed the triggers for its machine learning moderation algorithm, as well as changing processes.

According to TikTok's transparency report, more than 37 million videos were removed in India, from which the app is currently banned after a late June decision by the government to bar access to apps of Chinese origin.

India also had far and away the most requests for removals of content due to legal reasons, with 79% of the 1,206 requests acceded to.

In the US, where TikTok has faced a month of legal and political wrangling over the app's future, 9.8 million videos were removed for violating the app's terms. There, 290 legal requests to remove content were received, of which 85% were deemed legitimate.

TikTok does not operate in China. Rather, its parent firm ByteDance operates a similar app called Douyin, so TikTok did not receive any requests for removal of content from China.

The app also received requests from governments across the globe to remove content in accordance with local laws. Here, Russia dominated, with 296 pieces of content removed or receiving restricted access across 259 separate accounts.

TikTok has recently been criticised for moderating its content to local norms, with the result that some LGBTQ+ friendly hashtags in languages like Russian were effectively shadowbanned on the app.

More than 10,600 pieces of copyrighted content were flagged for removal, and nine in 10 were taken down by TikTok – nearly 10 times the number of notices received in the second half of 2019.