Powder Keg Technologies Selected for NEDO's NEP "Leap Course 3000" to Develop Diagnostic Technology That Continuously Verifies the Integrity and Robustness of Physical AI

Technology to diagnose robots end to end, from the communication infrastructure to the AI model, within a closed network. Security verification for physical AI, toward its safe deployment in society.

Powder Keg Technologies has been selected for the "Leap Course 3000" (Yakushin Course 3000) of the fiscal 2026 Program for Talent Discovery and Entrepreneur Development in Deep Tech (NEP), run by Japan's New Energy and Industrial Technology Development Organization (NEDO).

About NEDO's NEP Leap Course 3000

NEP (NEDO Entrepreneurs Program) is a NEDO program that discovers talent in deep tech fields and trains entrepreneurs. By creating and nurturing deep-tech startups, it aims to energize the economy and generate new industries and jobs.

The Leap Course 3000 supports eligible costs of up to JPY 30 million, with a subsidy rate of 1/1 (100%) and a project period of up to 12 months, from a date designated by NEDO through August 31, 2027. Across the entire fiscal 2026 Leap Course, 33 of 241 applicants have been selected as prospective subsidy recipients. For details, see NEDO's call for proposals for the FY2026 NEP Leap Course and its announcement on the implementation structure for the FY2026 NEP Leap Course (both in Japanese).

Background

Physical AI, in which AI mounted on a robot recognizes its surroundings and decides how to act on its own, is moving toward practical use. At the same time, it brings new security risks unique to robots.

The nature of the damage from a cyberattack depends heavily on what is attacked. In IT (information systems) such as office PCs and business systems, the assets to protect are data and information, and typical damage is data leakage or tampering. In OT (operational technology, meaning factory control systems), which covers production lines, PLCs (dedicated computers that control factory equipment) and industrial robots, the assets to protect are production equipment and continuity of operations, and typical damage appears as production line stoppages or equipment damage.

With robots equipped with physical AI, by contrast, whether industrial or autonomous mobile, what must be protected extends to human bodies and lives. If an attacked robot behaves unexpectedly, it could collide with people nearby and cause injury, and lost physical function or life cannot be restored.

Yet methods that diagnose a robot's communication infrastructure, its onboard AI model, and the factory network the robot connects to in one consistent process remain limited. Until now, we have developed and provided diagnostic technology for IT systems and OT environments through MUSHIKAGO, our AI-powered automated security assessment device. In this project, we will extend that expertise to robotics and physical AI.

What We Will Develop

In this project, "Research and Development of Diagnostic Technology for Continuously Verifying the Integrity and Robustness of Physical AI," we will develop diagnostic technology for robots running ROS and ROS 2. The diagnostics cover the robot's communication infrastructure, its AI model, and the entire network the robot connects to. Diagnosis will be completed within a closed network, so that it can be used in highly confidential development environments and at production sites. The work is organized around three themes.

1. Security diagnosis of robot control middleware

We will develop technology that automatically tests whether encryption and authentication work correctly in communication inside a robot and between robots. We will also visualize how far an intrusion originating from a robot could spread through the factory network (the expected scope of damage).

2. Security diagnosis of AI models on robots

We will develop technology to evaluate the resistance to attacks of the AI models responsible for a robot's perception and decision-making. By measuring robustness to small changes in input and the reliability of inference results, we aim to present an AI model's latent risks as a quantitative score. A distinguishing feature is that the evaluation method targets AI running on the robot itself (edge AI), where computing resources are limited, rather than large-scale AI in the cloud.

3. Combined system and AI diagnosis that keeps robots running

We will establish methods to carry out the two diagnoses above in combination, safely, on robots in operation. We will also research and develop "safety guardrails" that allow diagnosis without stopping the robot. This aims to identify vulnerabilities that span the communication infrastructure and the AI model, while preventing diagnosis from halting the robot or causing physical damage.

We plan to integrate these results as features of MUSHIKAGO. Our goal is an environment in which robot developers can verify robot security within their pre-shipment quality assurance process, without specialized security expertise.

Note: The development described here is based on the project's R&D plan. Product specifications, availability timing and sales terms will be announced as development and verification progress.

Outlook

We will first establish the diagnostic technology in an in-house environment using test robots, and then confirm its practicality through demonstrations in the real environments of Japanese robot manufacturers. After the subsidized project period ends, we aim to offer it in appliance (dedicated device) form to Japanese industrial robot manufacturers and developers of autonomous mobile robots, to support their pre-shipment security verification.

In the longer term, building on deployments in Japan, we are looking toward overseas markets, starting with Europe, and plan to develop the technology into a diagnostic platform that helps companies comply with the regulations of each country and region.

We will apply the automated diagnostic technology we have built in IT/OT security to the safe deployment of robots and physical AI, and contribute to an environment where Japan's robotics industry can take on the global market with confidence.

Contact

To take part in joint demonstrations related to this project, or to consult us about robot and AI security diagnostics, please use our contact form or email support@powderkegtech.com.