14  Frontier

15 Frontier

Recent work (roughly the last 12 months), newest first. Entries graduate into a topic chapter’s chronological timeline (Part II) once superseded or matured — keeping the book continuously evolving.

15.1 Signposts

Fast-moving paths shaping the next few years, each headed for a topic chapter.

• Recursive self-improvement / automated AI R&D — AI accelerating AI compresses oversight time and amplifies misalignment; the headline reason scalable oversight and capability thresholds exist. → Monitoring & Oversight
• AI control — assume a model may be scheming and design protocols safe regardless (Greenblatt et al., 2023). → Monitoring & Oversight
• Deceptive alignment / scheming — misaligned behavior that survives safety training (Hubinger et al., 2024). → Alignment
• Dangerous-capability evaluations — CBRN, cyber-offense, autonomy, self-proliferation; the trigger behind frontier safety frameworks. → Evaluation
• Safety cases — structured arguments that a system is safe enough to deploy (Hilton et al., 2025). → Systemic Safety & Governance
• Chain-of-thought monitorability — keeping reasoning legible for oversight, and the risk that optimization erodes it. → Interpretability

15.2 Intake queue

Dated entries below, newest first — one-line claim, why it matters, target topic. (Empty — populated during the continuous literature scan.)

Greenblatt, R., Shlegeris, B., Sachan, K., & Roger, F. (2023). AI control: Improving safety despite intentional subversion. arXiv Preprint arXiv:2312.06942.

Hilton, B., Buhl, M. D., Korbak, T., & Irving, G. (2025). Safety cases: A scalable approach to frontier AI safety. arXiv Preprint arXiv:2503.04744.

Hubinger, E., Denison, C., Mu, J., Lambert, M., Tong, M., MacDiarmid, M., et al. (2024). Sleeper agents: Training deceptive LLMs that persist through safety training. arXiv Preprint arXiv:2401.05566.