Security researcher Jun Song declined lucrative offers from a black hat group to train open-weight models on exploit data, cautioning that others may accept similar offers.
Security & Exploits ·
Security researcher Jun Song received substantial financial offers from a black hat group to train open-weight models on exploit data, but declined the proposals. Song has cautioned that others in similar positions may prove willing to accept comparable offers, raising concern about the potential weaponization of machine-learning systems against crypto infrastructure.
The incident highlights an emerging threat vector in which attackers seek to operationalize vulnerability discovery through AI trained on exploit datasets. Open-weight models, once trained on such data, could be distributed widely and used to accelerate the identification and execution of attacks against smart contracts, bridges, and other on-chain systems at scale.
It remains unclear whether other researchers or engineers have accepted such offers, what specific compensation amounts were involved, or how widespread these recruitment efforts by black hat groups may be. The precise mechanics of how exploit-trained models would be deployed—whether for automated scanning, attack generation, or some other purpose—are also not specified.