Google Colab Bans Deepfake AI Training to Curb Platform Abuse
Google updates its Colaboratory terms of service to prohibit the training of deepfake-generating AI systems. The new enforcement relies on a blacklist approach rather than broad keyword filters.
Google bans the training of deepfake-generating AI systems on its Google Colaboratory platform. The company updates its terms of use to explicitly include deepfake-related work on the list of disallowed projects, a change discovered by users running the open-source DeepFaceLab tool. Colab provides free and paid access to powerful hardware like GPUs and TPUs, making it a popular hub for AI researchers and developers.
Users attempting to run certain deepfake applications now receive error messages warning them about executing disallowed code. However, the enforcement appears to rely on a blacklist method rather than a sweeping keyword filter, as some popular deepfake projects like FaceSwap remain fully functional. This inconsistency suggests that Google relies partly on community reporting to identify and block prohibited code.
A Google spokesperson confirms that deepfakes were added to the prohibited activities list last month as part of ongoing reviews of abusive platform patterns. The company states it continuously monitors Colab for abuse that contradicts its AI principles while attempting to balance these restrictions with its mission to provide valuable computing resources to the public.