Research Lead, Pre-training Safety

Lead research on removing harmful capabilities from AI models during pre-training to prevent misuse and loss-of-control risks.
- Direct pre-training safety research at scale (>100B parameters), partnering with red teams to stress-test and validate capability-control methods.
- Build and mentor a technical team, set research direction, and remain hands-on with coding and experiments.
- Develop and publish novel approaches to data filtering, gradient routing, unlearning, and synthetic data integration for safer model training.
- Requires strong research track record in AI with deep experience in language-model pretraining, dataset construction, and large-scale training pipelines.
- Requires team leadership experience or mentorship of junior researchers; established AI safety publication record preferred.

Lead research on removing harmful capabilities from AI models during pre-training to prevent misuse and loss-of-control risks.
- Direct pre-training safety research at scale (>100B parameters), partnering with red teams to stress-test and validate capability-control methods.
- Build and mentor a technical team, set research direction, and remain hands-on with coding and experiments.
- Develop and publish novel approaches to data filtering, gradient routing, unlearning, and synthetic data integration for safer model training.
- Requires strong research track record in AI with deep experience in language-model pretraining, dataset construction, and large-scale training pipelines.
- Requires team leadership experience or mentorship of junior researchers; established AI safety publication record preferred.

FAR.AI is an AI safety research nonprofit focused on addressing risks from advanced AI systems.