ByteDance’s Doubao Under Fire for AI-Generated Deepfake Pornography

The rapid proliferation of generative artificial intelligence in China has encountered a severe ethical and security crisis as ByteDance’s flagship AI chatbot, Doubao, faces allegations of facilitating 'digital public shaming.' An investigation by the feminist media collective Free Nora has revealed that the platform is being systematically exploited to generate non-consensual pornographic images of real women. This development underscores the growing tension between the explosive growth of AI consumer tools and the lagging efficacy of safety guardrails in one of the world's most aggressive AI markets.

Doubao’s market dominance significantly amplifies the scale of this issue. According to data from QuestMobile, Doubao commanded 155 million weekly active users as of late December 2025, nearly double that of its closest rival, DeepSeek, which recorded 81.6 million users. This massive user base provides a vast surface area for misuse, particularly when platform moderation is perceived as 'weak' or easily circumvented. The investigation highlights that for many women, their digital likeness has been transformed into 'raw material' that can be extracted and sexually degraded at will, creating a new frontier of online harassment that is difficult to police.

“

According to data from QuestMobile, Doubao commanded 155 million weekly active users as of late December 2025, nearly double that of its closest rival, DeepSeek, which recorded 81.6 million users.

Technically, the abuse is fueled by a sophisticated system of 'jailbreaking' known as fenjue. Named after a secret technique from a Chinese fantasy novel, fenjue consists of a coded system of prompts designed specifically to bypass the safety filters of large language models (LLMs). Free Nora’s volunteers spent months infiltrating anonymous groups on Telegram, where users collectively fine-tuned these methods to trick Doubao’s algorithms into generating explicit content. This collaborative effort to undermine AI safety suggests that static moderation filters are increasingly inadequate against a motivated and organized user base that treats prompt engineering as a tool for exploitation.

While China has been a global leader in drafting AI-specific regulations—including rules on deep synthesis and generative AI services—the Free Nora report suggests a significant gap between policy and practice. The Cyberspace Administration of China (CAC) has previously mandated that AI providers must implement robust content moderation and verify user identities. However, the ease with which Doubao’s safeguards were bypassed indicates that enforcement remains inconsistent. This regulatory lag is not unique to China, as deepfake pornography remains a global concern, but the sheer scale of Doubao’s reach makes the lack of effective guardrails particularly damaging to public trust.