Sakana AI Launches Fugu-Cyber

Sakana AI has unveiled Fugu-Cyber, a new orchestration model purpose-built for advanced cybersecurity operations. This update to their Fugu model aims to tackle the intricate demands of modern cyber defense.

Fugu-Cyber demonstrates state-of-the-art performance on critical security benchmarks. It achieved an 86.9% success rate on CyberGym, which assesses code vulnerability analysis, and 72.1% on CTI-REALM, measuring threat intelligence translation. These figures rival leading cybersecurity-focused models like GPT-5.5-Cyber.

Like its predecessor, Fugu-Cyber operates as a multi-agent system that presents a unified interface. Users interact with a single API endpoint, while the system intelligently coordinates specialized agents for complex, multi-step tasks. This design mitigates risks associated with single-vendor dependencies.

The Reality of Frontier Cyber Capabilities

Sakana AI cautions against the hype surrounding frontier models in cybersecurity. Simply providing access to advanced AI, even with cyber capabilities, does not automatically solve an organization's security challenges.

Real-world deployment requires more than just raw AI power. Large enterprises often struggle to integrate these tools effectively without specialized internal talent and deep understanding of proprietary systems.

This sentiment is echoed in industry reports and Sakana AI's own experience working with major Japanese institutions.

Operationalizing AI for cyber defense demands a fusion of human cybersecurity expertise and advanced AI capabilities. A powerful API is a crucial component, but not the entire solution.

Beyond the API: Enterprise-Grade Security

Raw AI models, when deployed in isolation, are prone to generating false positives and lack the context for live production environments. True enterprise defense necessitates rigorous verification workflows.

Sakana AI's approach involves validating AI-identified vulnerabilities through specialized sub-agents and human-in-the-loop processes. This ensures proposed patches address actual, exploitable risks.

The company's Applied Enterprise team is developing the necessary infrastructure to deploy Fugu-Cyber safely. They are building specialized harnesses and workflows to integrate these models into production environments.

By combining Fugu-Cyber's reasoning with human security expertise, Sakana AI enables automated vulnerability verification and related tasks that are both powerful and reliable.

This orchestration of multiple AI models aligns with industry trends pointing towards future cyber defense systems that leverage collective AI performance for superior results.

Responsible Deployment

Given the sensitive nature of Fugu-Cyber's applications, Sakana AI is implementing a responsible release strategy. An updated Acceptable Usage Policy prohibits misuse and adheres to industry safety standards.

Access to the Fugu-Cyber API will be managed through a Token Plan, requiring users to submit an access request form. Each application will undergo manual review to ensure intended use cases align with security objectives.

Sakana AI is committed to ongoing system vetting and collaboration with enterprise partners to ensure this technology fortifies critical infrastructure.

Sakana AI, the company behind the Fugu orchestration models, is actively seeking talent to join their mission. Interested candidates can explore career opportunities.