Sakana AI Defense and Intelligence Team Places 5th in Diver OSINT CTF 2026 Using Fugu Agent
Sakana AI ● Covered by 2 sources
Sakana AI's defense team just placed 5th out of 850+ teams in an OSINT competition using their own AI agent. They paired human intel experts with an AI called Fugu-ultra 1.1, and it actually worked.
Based on reporting by Sakana AI — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Sakana AI, the Tokyo-based lab known for its evolutionary approach to AI models, sent a volunteer team from its defense and intelligence group into the Diver OSINT CTF 2026, a competition built around open-source intelligence gathering and analysis. Out of more than 850 competing teams, Sakana's group finished 5th. That's not a headline result in the way a chart-topping benchmark score is, but it's a meaningful data point for anyone tracking how AI agents perform in messy, real-world analytical tasks rather than clean academic tests.
The team didn't just throw an AI at the problem and hope for the best. Domain experts in intelligence work paired up with engineers to build an OSINT-focused agent powered by Fugu-ultra 1.1, Sakana's own model. The pairing matters: OSINT competitions reward the kind of pattern-spotting and source-verification instincts that come from actual tradecraft experience, not just raw computational horsepower. Sakana structured the effort so the agent handled scale and speed while humans supplied judgment and domain framing.
A 5th-place finish among 850-plus teams is a solid showing, and Sakana is framing it less as a trophy and more as validation of a working method. The company's pitch is that combining human expertise with AI agents isn't just efficient, it's specifically applicable to defense and security contexts where stakes are higher and errors more costly. That's a notable claim coming from a company, since plenty of AI vendors talk about human-AI collaboration in the abstract without ever testing it against outside competitors under time pressure.
Sakana says it plans to take what it learned from the competition and refine the agent for operational use, with an eye toward contributing to Japan's defense and intelligence capabilities. Given Japan's increasing interest in domestic AI capacity for security applications, a lab like Sakana positioning itself in this space is worth watching, even if the near-term output is a competition placement rather than a fielded system.
My take — AI-written commentary, not fact-checked reporting
I like that Sakana tested this in a public competition instead of just publishing a paper claiming their agent is great at intelligence work — a 5th place finish against 850 teams is a real signal, not marketing copy. That said, OSINT competitions and actual defense intelligence operations are different beasts, and I'd want to see how Fugu-ultra holds up when the data is deliberately deceptive rather than just abundant. Still, more labs should be benchmarking security-relevant AI this way rather than hiding behind vague safety claims.
Read more about this at: Sakana AI
