Sakana AI says Japan should chase AI sovereignty through post-training, not frontier pretraining

SakanaAILabs · x · 2026-07-29

Sakana AI says post-training, not pre-training, is Japan’s path to AI sovereignty

The company’s defense chief argues that Japan cannot realistically catch up with U.S. and Chinese frontier-model development by spending huge sums on pre-training. Instead, Sakana AI is focusing on post-training: adapting existing open models to local security, policy, and factuality requirements without sacrificing performance.

Sakana AI has entered defense in 2025 and is working on AI for command-and-control systems and misinformation analysis.
The company says its role is to support both Japan’s Ministry of Defense and prime contractors.
It frames defense as a software-first market where AI can shape strategy and tactics, not just hardware performance.
The interview also highlights a broader thesis: Japan should secure sovereign capability in model adaptation and bias reduction, rather than trying to beat frontier labs at base-model pretraining.

Original post →

Sakana AI says Japan should chase AI sovereignty through post-training, not frontier pretraining

More from Companies & People