We introduce provable and practical watermarking approaches for data poisoning attacks.