AI Explainability and Deep Learning Techniques for Security: JD Security’s Recent Research Highlights

AI Explainability Technology Makes Security Traceable

After AI applications entered the security field, a fundamental dilemma emerged: whether AI decisions are scientifically justified or merely a gamble, especially given the irreversible nature of security outcomes.

In e‑commerce, JD Security experts argue that freezing accounts or labeling software as malicious cannot rely solely on opaque machine decisions. To address this, JD Security Lab partnered with Professor Xing Xinyu’s team at Pennsylvania State University to develop an “AI parsing” technique.

This technique leverages Gaussian‑model approximations to provide accurate explanations for any deep‑learning model, ensuring that AI‑driven anti‑fraud actions are well‑justified and minimizing false positives against legitimate users. It also enables automatic patch generation and self‑repair of AI security systems.

Deep Learning for Program Crash Analysis Brings AI into Traditional Security

Server crashes generate core dumps containing memory states and executed instructions. JD Security applies deep‑learning models to these dumps, reconstructing the program’s state before failure and automatically identifying root causes, thereby dramatically reducing the time and manpower required for bug fixing.

This approach lowers both human and financial costs of crash response while preserving user experience.

AI‑vs‑AI: The Ongoing E‑Commerce Attack‑Defense Arms Race

With millions involved in cyber‑crime, AI‑empowered black‑market operations have become highly organized and intelligent. JD Security combats this through multi‑layered strategies: anti‑scraping measures, AI‑driven detection of automated registrations using unsupervised and supervised learning, and advanced human‑machine verification.

For captcha challenges, JD Security extracts the opponent’s AI model, generates adversarial samples that remain readable to humans but confuse malicious AI, thereby degrading the effectiveness of black‑market bots.

Open‑Source AI Collaboration Builds a Borderless Security Community

Recognizing that many enterprises lack the data and resources to conduct extensive AI security research, JD Security advocates for open‑source sharing of AI security tools to foster a global, collaborative defense ecosystem.

Chief Security Scientist Tony Lee emphasizes the resurgence of the “white‑hat” spirit and calls for a worldwide “borderless doctor” network to collectively eradicate security poverty.