Bottlenecks for defense AI: cost, dual-use risk, and willingness, per HF researcher

eliebakouch · x · 2026-09-13

Hugging Face researcher Elie Bakouch lists the main bottlenecks for applying AI in defense: cost, dual-use risk, time, and the willingness of software owners to cooperate. His proposals: get as many actors as possible to partner, extend initiatives like Glasswing to more organizations, massively subsidize inference tokens for defense actors, and offer technical support. He notes cheap, capable open-source models like DeepSeek V4.1 Flash could already help if properly orchestrated.

Related event: What Blocks Frontier Labs from Securing the World's Software?(2 posts)→

Original post →

More from Safety

Safety channel →