#USA #Air Force #AI Detection #Stream Engine #Rome, NY

Stream Engine and AFRL Collaboration

Stream Engine, a leading provider in decision intelligence for national security, recently announced a significant milestone in its collaboration with the Air Force Research Laboratory (AFRL). Under a Cooperative Research and Development Agreement (CRADA), Stream Engine has successfully developed a cutting-edge, containerized semantic-based deepfake detection analytic. This tool is designed to enhance the capabilities of the AFRL's Information Directorate, marking a crucial advancement in national security technology.

Rapid Development to Test-Ready Status

The development phase of this project was impressively swift. According to the annual report by AFRL, the deepfake detection analytic was integrated into the lab's operational Test and Evaluation testbed and was officially ready for testing by June 2026. This accomplishment reflects both organizations' commitment to innovation in security technology and emphasizes the importance of real-time solutions for identifying threats in the information landscape.

The CRADA agreement was executed on June 6, 2025, with a duration of 12 months, aimed at integrating and evaluating advanced detection technology. As part of this collaboration, Stream Engine built the analytic from the ground up, while AFRL focused on optimizing its testbed to evaluate the system's performance under various conditions. The combined efforts are anticipated to yield vital insights into the efficacy of deepfake detection methods in real-world applications.

Advanced Detection Techniques

Stream Engine applied its expertise in large language models (LLM) to devise a two-stage deepfake detection pipeline. This innovative methodology begins with a semantic feature extraction model that scrutinizes textual content for grammatical and syntactic anomalies as well as any semantic inconsistencies. Subsequently, a neural network classifier synthesizes these features into likelihood scores, effectively distinguishing between machine-generated text and poorly written human content. This dual approach enhances the reliability of deepfake detection beyond surface-level analyses, setting a new standard for this critical area of security.

To streamline deployment, the analytic was packaged as a container utilizing an AFRL-engineered Docker template. This approach facilitated secure integration into the testbed environment. In addition, AFRL provided essential synthetic and manipulated datasets for the training phase, along with a tailored software environment for integration and testing.

Overcoming Challenges to Success

Despite facing significant obstacles, including a facility power outage and a six-week federal government shutdown during the development phase, the team overcame these challenges and achieved test-ready status on schedule. The AFRL has confirmed that the Test and Evaluation environment is configured for current and future projects, underscoring the robustness of the analytic and the collaborative effort.

No federal funds were exchanged during this collaboration; however, Stream Engine's contributions are valued at more than $1 million. This estimation covers personnel, intellectual property, supplies, and equipment that were integral to the development and delivery of the deepfake detection analytic.

Insights from Stream Engine's Leadership

Brian Kim, co-founder and CEO of Stream Engine, shared insights regarding the significance of this project: "We were tasked with utilizing our language intelligence capabilities to tackle one of the most challenging problems in national security – distinguishing between content generated by humans and that produced by machines. The analytic was developed from the ground up and is now positioned in AFRL's testbed, ready for formal evaluation. This rapid development reflects the urgency required in security missions and epitomizes what Stream Engine strives to accomplish: transforming signals into actionable decision intelligence for those safeguarding the nation."

The Future of the Collaboration

With the testbed now fully configured, the next phase of this collaboration involves formal testing and evaluation of the developed analytic, promising to advance the field of deepfake detection and contribute to national security efforts. Stream Engine's commitment to harnessing AI for good continues, as it supports defense and intelligence organizations in navigating the complexities of modern security challenges.

About Stream Engine

Stream Engine stands at the forefront of decision intelligence technology for national security. Its AI platform enhances situational awareness by converting text, images, audio, and video data from various sources into prioritized intelligence, empowering security teams to respond more effectively to emerging threats. From combating cybercrime to enhancing safety measures at large events, Stream Engine is dedicated to leveraging artificial intelligence for a safer future. For more information, visit stream-engine.io.