Watermarking enables language model providers to identify text generated by their models. However, its desired properties can conflict (\ie~stronger watermark signals can degrade text quality), while designs that resist editing may also facilitate forgery. Providers address these trade-offs by choosing configurations t …
Generative models can synthesize high-quality inauthentic multimedia content that is already being misused at scale. We evaluate twenty deepfake detectors against ten generators released in the last four years and find accuracy decreasing over time, from near-perfect 99.5% to 76%. Adversarial perturbations further redu …
Multimodal encoders such as CLIP underlie many downstream systems, but their web-scale training data embed harmful associations that safety alignment must suppress without unnecessarily changing benign representations. Because ethical and practical constraints prevent collecting real unsafe content at scale, existing d …