Jun Song says anonymous black-hat group offered high payment for exploit-data model training
Open-source AI practitioner Jun Song said in a post on X that an anonymous black-hat group contacted him yesterday with a reinforcement learning request and offered what he described as a very high payment. According to Song, the group wanted him to carry out post-training on an open-weight model using exploit data. He said the scale of the attack data the group claimed to hold was unlike anything he had seen before, adding that if someone told him it was an undercover law-enforcement dataset, he would believe it. Song also said he runs a data center hardware export business in South Korea and is advancing the 0xSupergemma project. As a comparison, he pointed to a white-hat project he worked on that used GLM-5.3; its dataset was less than 1% of the size described by the group, yet its performance already far exceeded frontier models. Song said he rejected the request, but added that his real concern is that someone else may eventually accept the deal.
