Sakana AI Recruits Former NSA and CIA Leaders for Defense Push

Tokyo-based Sakana AI has appointed retired U.S. Navy Admiral Michael Rogers and former CIA Chief of Station Tom Moran as Senior Advisors focused on defense and cybersecurity, according to the company’s announcement.

The appointments follow Sakana’s contracts with Japan’s Ministry of Defense and its Acquisition, Technology & Logistics Agency. They give the AI lab access to senior operational experience as it develops models for intelligence analysis, edge devices, command systems, and cyber defense.

Operational résumés, narrow briefs

Rogers led the National Security Agency and U.S. Cyber Command from 2014 to 2018, overseeing signals intelligence and military cyber operations. He retired as a four-star admiral after 37 years in the Navy and now serves as a senior adviser at Brunswick Group.

Moran spent much of his CIA career in human intelligence, serving six times as Chief of Station across East Asia, Western Europe, the Middle East, and Africa. He also held senior positions at the agency’s headquarters and directs the Intelligence Project at UC Berkeley’s Goldman School of Public Policy.

As Senior Advisors, Rogers and Moran have a defined subject-matter role. Sakana has disclosed no executive authority, board vote, compensation, or time commitment for either appointment.

Contracts point to the work

Sakana’s agreements with Japanese defense agencies explain the emphasis on operational intelligence and cybersecurity. The projects move its research into systems that must work with limited computing power, sensitive data, unreliable network connections, and strict audit requirements.

The ATLA project centers on vision-language models, which process images and text together. Smaller versions can run close to where data is collected, reducing bandwidth use and allowing faster analysis when a device cannot maintain a reliable connection to a data center.

The ministry project applies AI agents to multi-step intelligence work. Such systems can gather material, compare sources, and draft reports, while analysts retain responsibility for source validation and consequential decisions. Reliable deployment will require traceable outputs, reproducible workflows, access controls, and clear human approval points.

Sakana’s publicly stated defense priorities now cover four connected areas:

• Modernizing command-and-control systems through multi-domain data fusion
• Running small vision-language models on drones and handheld devices
• Using AI agents to support intelligence analysis and strategic decisions
• Strengthening cybersecurity across models, data, and deployment infrastructure

Japan funds a sovereign stack

Japan’s rising defense spending and its focus on “information power” have created new procurement opportunities for domestic AI companies. National-security buyers increasingly seek local control over model weights, adaptation data, inference infrastructure, and software supply chains.

Sovereign AI describes this effort to keep strategic computing capacity and sensitive data under national or domestic control. Achieving that goal requires more than hosting a model locally. Agencies also need verified model components, data-residency controls, security testing, incident response, audit logs, and procedures for accrediting systems on protected networks.

Sakana’s evolution-inspired research fits part of that demand. Its methods search for effective ways to combine or adapt existing models, including through model merging. This approach could produce specialized systems without the cost of training a frontier model from the beginning, although each merged system still requires task-specific safety, security, and performance evaluations.

Advisers arrive as Sakana scales

Sakana has named defense and finance as priority markets while expanding from the combination of existing models toward proprietary model development. The company has also appointed Goldman Sachs as an adviser and begun preparations for a potential initial public offering, though it has announced no listing date.

The earlier appointment of AI researcher Jürgen Schmidhuber broadened Sakana’s research bench. Adding Rogers and Moran extends that advisory roster into cyber command, intelligence collection, and government operations. Their backgrounds may inform product requirements, procurement strategy, risk controls, and discussions with regulators and public-sector customers.

The appointments provide no evidence by themselves that Sakana’s systems meet military deployment standards or can process classified information. Those conclusions will depend on technical evaluations, security accreditation, contract milestones, and field performance.

Signals developers can measure

Engineering teams pursuing Japanese government and regulated-enterprise work can expect requirements similar to those reflected in Sakana’s roadmap:

• Constrained inference: Multimodal models must run within tight limits on memory, power, bandwidth, and latency.
• Traceable agents: Automated research workflows need source citations, durable logs, reproducible steps, and human review.
• Secure integration: Models must connect with existing command, identity, data-fusion, and reporting systems without weakening access controls.
• Supply-chain assurance: Buyers will scrutinize model origins, dependencies, training data, update mechanisms, and third-party components.
• Operational evaluation: Accuracy tests must cover adversarial inputs, degraded networks, unfamiliar environments, and costly false positives.

Future contract expansions, deployment results, security certifications, and product disclosures will show how much influence the new advisers have. Those milestones will also indicate whether Sakana can convert its research methods and government relationships into dependable defense systems.