July 21, 2026
Today we are releasing an update to our Fugu orchestration model: Fugu Cyber.
Available as a new API endpoint, Fugu-Cyber is purpose-built to address the complexities of modern cyber defense. Fugu-Cyber achieves top performance on the industry’s most demanding security benchmarks, achieving a pass rate of 86.9% on CyberGym and 72.1% on CTI-REALM, comparable to leading cybersecurity-focused frontier models such as GPT-5.5-Cyber and Mythos-Preview.
Fugu-Cyber achieves top performance in real-world security benchmarks and matches cyber-focused frontier models such as GPT-5.5-Cyber and Mythos Preview.
Together, these benchmarks test the cornerstones of enterprise defense: CyberGym Evaluates an agent’s ability to analyze complex code bases to verify real-world vulnerabilities CTI REALM measures its ability to translate raw threat intelligence reports into working detection rules.
Like our original Fugu orchestration model, Fugu-Cyber is a multi-agent system that behaves like a single model. You send a request to an endpoint and the system dynamically orchestrates a pool of specialized agents to handle complex, multi-step tasks without the risk of dependency on a single provider. It is now available as a new API endpoint in sakana.ai/fugu
Table of Contents
The reality check on frontier cyber capabilities
Achieving high scores on an assessment is just the beginning of the story.
There has been a lot of fear lately about the cyber capabilities of frontier models. Many industry narratives suggest that simply giving them access to a frontier model with cyber capabilities will immediately solve an organization’s security challenges.
We believe it is time to turn this conversation into reality.
As highlighted in a recent Nikkei Digital Governance report (in Japanese), easy access to a boundary model like Anthropic’s Mythos is not the magical solution to enterprise security. In reality, large organizations, including large financial institutions, often struggle to implement these tools. Without specialized in-house talent and deep integration with proprietary source code, a frontier model, even with state-of-the-art cyber capabilities, cannot easily uncover or remediate real-world vulnerabilities.
The challenges highlighted in the Nikkei article also reflect Sakana AI’s own experiences working with the largest Japanese companies to address cybersecurity challenges. Successful deployments require both human Cybersecurity expertise and access to border functions.
A high-performance API with strong cyber reasoning is an incredibly important piece of the puzzle. It is not the complete solution.
Beyond the API
When deployed in isolation, raw models inevitably lead to false positives. Without the right equipment, they will have a hard time understanding the nuances of a live production environment. True enterprise defense requires more than just access to a frontier model.
In our experience, cybersecurity solutions require the use of boundary models with deep, localized human cybersecurity expertise and rigorous verification workflows. When an AI system discovers a potential vulnerability, it must be validated by subagents specialized in cybersecurity and human-in-the-loop processes to confirm whether it would actually trigger in a real-world environment before proposing a patch.
This is exactly the challenge that Sakana AI solves.
The Enterprise Solution: Bridging the Gap Between Fugu-Cyber’s Border Functions and Enterprise Security
This is where Sakana AI’s Applied Enterprise team comes into play. We don’t just build the core engine; We are building the infrastructure required for secure use.
We are currently working closely with major Japanese institutions to develop the specialized wiring harnesses and workflows required to use these models in production. By combining Fugu-Cyber’s raw reasoning power with the real-world experience of security experts, we help companies create automated vulnerability scanning and other follow-up tasks that are both high-performance and highly reliable.
As industry experts have noted, the future of cyber defense depends on systems that can combine multiple AI models to achieve higher performance than any single model could alone. By orchestrating the world’s best models into a unified system, we deliver the realistic, resilient blueprint necessary for AI sovereignty.
Responsible use
Since Fugu-Cyber deals with sensitive security processes, we are committed to their safe and responsible publication.
To ensure safe delivery, Fugu-Cyber is released under an updated Acceptable Use Policy that prohibits offensive abuse and aligns with industry security standards.
The Fugu Cyber API will be available for Token plan.
Users must also request access by submitting an access request form detailing their intended use case and providing verified contact information. Our team carefully reviews and approves each application manually before granting users access to Fugu-Cyber.
We will continue to review our systems and work with our corporate partners to ensure this technology is used to strengthen and defend critical infrastructure.
Sakana Fugu-Cyber is available today as a new model on our API endpoint at sakana.ai/fugu.
To find out more about our business solutions, please contact our Applied team.
Section AI
Are you interested in joining us?
For more information, see our career opportunities.
https://sakana.ai/fugu-cyber-release/
