TikTok Announces AI System Automatically Deletes 77% of Harmful Content, Enhancing User Safety
Daniel Kim Views
TikTok is actively working to improve user safety by using artificial intelligence (AI) to filter harmful content. The company employs AI to detect and remove offensive videos automatically and has a global team of 40,000 people to review and further filter videos according to local guidelines and sensitivities.
On August 1st, TikTok held a global press conference at its Transparency & Accountability Center (TAC) in Singapore to introduce these user protection measures.
The TAC was established to identify harmful content and create a safer platform environment. In Los Angeles in 2020, TAC centers opened in Dublin (2023), Singapore, and Washington DC (May 2024).
Inside the Singapore TAC, there’s an AI demonstration area showcasing how the system detects harmful content. This AI system identifies and flags content related to dangerous tools, alcohol, extremism, and smoking by analyzing user behavior. For instance, if a user simply holds a drink, the AI assesses it as low risk. However, if the user drinks from it, the risk level increases. Similarly, holding a knife is considered low risk, but bringing it close to the neck significantly raises the danger level.
Every day, TikTok’s AI removes 2 million harmful videos. The AI system continuously improves through machine learning by analyzing daily uploaded videos.
To ensure no harmful content slips through, TikTok employs 40,000 content moderators worldwide. These moderators understand the local markets and remove videos that conflict with social and cultural norms. Overall, 77% of harmful content is filtered by AI, with the rest handled by human moderators.
KahnJi, TikTok’s Outreach & Partnership Manager, stated that in the first quarter of this year, 97.7% of harmful videos were removed by AI before being reported. He noted that the removed content represents less than 0.9% of all uploaded videos, emphasizing the company’s ongoing commitment to being a safe and responsible platform.
Most Commented