Could (/should) this be mitigated on the repository host side by scanning for and flagging malicious commits? A paper from last year evaluating this on Github

Could (/should) this be mitigated on the repository host side by scanning for and flagging malicious commits? A paper from last year evaluating this on Github achieves a ~50% success rate[0]. Given Github already training ml models across all repositories for Copilot I would guess higher rates would be possible. [0] https://arxiv.org/abs/2103.03846 https://arxiv.org/abs/2103.03846 Edit: add link