Miles Jamison 
- CTC and the Robotics Research Center are jointly developing the AI Security Assessment Tool
- The tool will use a model-stealing pipeline to test machine learning vulnerabilities
- The pipeline will assess risks involving privacy leaks and adversarial attacks on AI systems
Concurrent Technologies Corp. has partnered with the U.S. Military Academy’s Robotics Research Center, or RRC, at West Point to build a machine learning tool meant to guard defense artificial intelligence systems against adversarial attacks and privacy vulnerabilities, the Johnstown, Pennsylvania-based company said Thursday.
What Does the Tool Do?
The initiative centers on building an AI Security Assessment Tool that uses a model-stealing pipeline to test vulnerabilities in machine learning systems. The pipeline is designed to flag which architectures and training methods hold up worst against privacy leaks and put safeguards like differential privacy and adversarial training, an area that former president and CEO Ed Sheehan Jr. previously identified as part of the company’s broader AI cybersecurity work, through their paces against attempts at theft.
Rather than building from scratch, the team assembled the pipeline out of codebases CTC had already written for earlier projects giving researchers a system they can reconfigure for different experiments going forward.
CTC President and CEO Chris Hamilton said the RRC collaboration underscores the company’s efforts to push AI security forward while supporting the military’s readiness and technological capability.
“As artificial intelligence becomes increasingly integrated into defense and national security applications, ensuring the security and integrity of these systems is paramount,” said Hamilton.
How Will the Technology Be Applied?
The pipeline is also meant to show how well certain privacy techniques hide the traits of training data from attackers and to pinpoint which parts of a model an attacker is most likely to target.
Dave Saranchak, executive director at CTC, said the company’s team contributed expertise toward building tools meant to guard machine learning models against adversarial attacks and unauthorized access, calling the work necessary to keep the AI systems the country relies on secure and intact.
