Executive Summary: As of May 2026, Chinese Advanced Persistent Threat (APT) group APT10 has refined its operations by integrating generative AI into open-source intelligence (OSINT) collection, enabling highly targeted attacks against Taiwan’s semiconductor supply chain. Leveraging GitHub—a critical hub for global developer collaboration—APT10 exploits AI-driven reconnaissance to identify vulnerable software dependencies, exposed credentials, and misconfigured repositories. These attacks are part of a broader geopolitical strategy to disrupt semiconductor manufacturing, a cornerstone of Taiwan’s economic and technological sovereignty. This article examines APT10’s evolving tactics, the role of AI in OSINT, and the vulnerabilities within the Taiwanese semiconductor ecosystem. Recommendations are provided for defenders to mitigate these advanced threats.
APT10, also known as MenuPass or Stone Panda, has long been associated with sophisticated cyber espionage campaigns targeting governments and critical infrastructure. By 2026, the group has integrated generative AI into its OSINT workflows, significantly enhancing its ability to gather, analyze, and act on publicly available data. Traditional OSINT techniques—such as scraping social media, analyzing leaked databases, and monitoring domain registration records—remain foundational. However, the addition of large language models (LLMs) and AI-driven natural language processing (NLP) has allowed APT10 to automate and scale its operations.
In particular, APT10 has developed custom AI tools to parse GitHub repositories, Git commit logs, and package management files (e.g., `package.json`, `requirements.txt`). These tools can identify:
For example, APT10’s AI models can scan thousands of Taiwanese semiconductor-related repositories in minutes, prioritizing those with high-impact dependencies or recent commits by key engineers. This intelligence is then used to craft highly tailored spear-phishing emails or to inject malicious code into seemingly legitimate projects.
Taiwan is home to some of the world’s most advanced semiconductor manufacturers, including TSMC, UMC, and MediaTek. These companies rely heavily on open-source software and collaborative development platforms like GitHub to accelerate innovation. However, this reliance introduces significant cybersecurity risks:
APT10’s AI-powered OSINT allows it to map these dependencies with precision. For instance, the group may identify a Taiwanese semiconductor firm using an outdated version of a critical OSS library. It can then craft a malicious pull request or commit that propagates malware through the dependency chain, affecting not only the target but also its downstream customers.
GitHub’s role as the de facto platform for software development makes it a prime target for APT10. The group exploits several GitHub-specific vulnerabilities:
APT10 leverages AI to identify repositories that rely on private or internal package registries but accidentally expose dependency lists in public files (e.g., `requirements.txt`). The group then uploads malicious packages with higher version numbers to public registries (e.g., PyPI, npm), which are automatically pulled in by unsuspecting developers.
By analyzing Git commit histories, APT10 can identify developers who recently modified critical files (e.g., firmware updates, configuration files). The group then impersonates these developers via compromised accounts or phishing attacks to inject malicious code.
APT10’s AI tools can detect misconfigured GitHub Actions workflows that allow untrusted code to execute with elevated privileges. These pipelines are often used to build and deploy semiconductor-related software, making them high-value targets for sabotage or espionage.
A notable 2025 incident involved APT10 compromising a CI/CD pipeline at a Taiwanese semiconductor equipment manufacturer. The group inserted a backdoor into firmware updates, which were then distributed to customers via automated updates. The malware remained undetected for weeks, exfiltrating proprietary design data.
APT10’s AI-driven OSINT operations are structured around three core capabilities:
The group uses custom-built AI agents to scrape and index data from GitHub, GitLab, Stack Overflow, and developer forums. These agents employ NLP to extract relevant information, such as:
APT10’s AI models analyze temporal patterns in developer activity to predict future changes or deployments. For example, if a semiconductor firm’s GitHub repository shows increased activity around a specific firmware release, APT10 may time its attack to coincide with that release.
The group uses generative AI to craft highly personalized phishing emails or messages. These may mimic internal communications from IT departments, project managers, or even open-source maintainers, increasing the likelihood of success.
In one documented case, APT10 impersonated a well-known open-source maintainer to trick a Taiwanese engineer into downloading a malicious npm package disguised as a legitimate utility for semiconductor design.
To counter APT10’s evolving tactics, organizations in Taiwan’s semiconductor supply chain must adopt a multi-layered defense strategy: