Recent advancements in AI have opened up intriguing opportunities, but also pose societal risks. One of our research goals is to mitigate the negative impact of AI, for example, by developing deepfake detection methods, accurately estimating prediction uncertainty, and ensuring user privacy.