返回新聞
安全性AI Understanding 簡報

GitHub AI代理識別24個Android應用程式漏洞

GitHub 安全实验室研究员 Kevin Stubbings 使用自定义 AI 驱动的审核工作流程发现并报告 Android 应用程序中的 20 多个漏洞,包括 OsmAnd 和 Wikipedia 中的关键缺陷。

4 min readRead the linked source
Source-provided image accompanying GitHub AI agent identifies 24 Android app vulnerabilities
來源參考來源記錄
出版商
helpnetsecurity.com
來源連結
helpnetsecurity.comhttps://www.helpnetsecurity.com/2026/09/29/github-ai-android-app-vulnerabilities/
來源類型
連結來源-主要來源狀態尚未確定。
背景60 秒內了解這一點

從這裡開始

關鍵術語

人工智慧代理
一種可以觀察、推理並採取行動來實現目標的軟體系統,通常使用工具和記憶體。
測試一下自己AI 代理測驗

發生了什麼事

GitHub Security Lab researcher Kevin Stubbings developed custom AI-driven audit workflows, known as taskflows, using the lab's open-source Taskflow Agent. These workflows were specifically designed for mobile app security, prompting the AI to check for intent-based bugs like confused deputy issues and insecure broadcasts. Using these tools, Stubbings identified and reported more than 20 vulnerabilities in Android applications, including significant security flaws in the OsmAnd navigation app and the Wikipedia Android app.

GitHub Security Lab researcher Kevin Stubbings built custom AI-driven audit workflows, called taskflows, on top of the lab’s open source Taskflow Agent. These workflows were tailored for mobile apps, adding specific steps to separate mobile entry points from web or desktop ones and prompting the model to check for intent-based bugs, such as confused deputy issues and insecure broadcasts, which generic security prompts tend to miss.

Using these taskflows, Stubbings found and reported more than 20 vulnerabilities in Android apps. Two notable examples include a flaw in OsmAnd, a navigation app with over 10 million downloads, where an exported activity accepted intent extras that should have been restricted. This allowed any app on the phone to silently import malicious settings, such as swapping the map tile source for an attacker-controlled server to log user coordinates. Another flaw in the Wikipedia Android app involved a hostname check in its deeplink handler that used endsWith() instead of matching the full domain, allowing lookalike domains to load in the app’s WebView and potentially steal session cookies.

來源詳情: helpnetsecurity.com ↗

為什麼這很重要

This development demonstrates the practical utility of AI agents in identifying complex, context-specific security vulnerabilities in mobile software that generic security prompts often miss. The findings highlight both the power and the limitations of current AI security tools, as the models excelled at finding bugs but struggled with accurately assessing their severity and real-world impact. This underscores the continued necessity for human expertise in security auditing, even as AI tools become more capable of automating the discovery phase of vulnerability research.

The incident illustrates the growing role of AI agents in software security, specifically in identifying nuanced vulnerabilities in mobile applications. By targeting specific mobile security patterns, the AI was able to uncover issues that might be overlooked by broader, less specialized security scans.

However, the source notes that the AI was better at finding bugs than judging their severity. The model frequently flagged low-severity issues and misjudged real-world impact when mitigating factors were present. This highlights a current limitation in AI security tools: they can automate discovery but still require human reviewers with domain knowledge to validate findings and assess true risk before remediation.

Interactive Mechanism

互動機制:它實際上是如何運作的

以互動方式探索這項發展背後的基礎技術。

Agent Lifecycle Stage:
1
User Intent & Planning: "Audit customer refund request #4092 and settle payment."
2
Tool Calling: Emits structured JSON call crm_get_transaction(id='4092').
3
Guardrail & Verification:🛡️ Paused: High-value action requires human operator sign-off.
4
Final Settlement: Refund recorded, email receipt dispatched, and audit log stored.
Core takeaway: An AI agent is not just a language model—it is a closed loop of planning, tool invocation, and environment feedback. Production systems require self-healing retries and strict human approval guardrails.
互動式概念檢查+10 Points
AI Agents Quiz

What most distinguishes an AI agent from a basic chatbot?

接下來看什麼

Monitor the adoption of AI-driven security audit workflows by other security labs and enterprises. Watch for updates on the open-source Taskflow Agent and whether similar AI tools are being integrated into standard CI/CD pipelines for mobile development. Additionally, observe how app developers respond to these AI-discovered vulnerabilities, particularly regarding the patching of intent-based and deeplink security flaws.

The taskflows are open source and free to run against any repository, though they require a GitHub Copilot license and can consume a significant number of premium model requests. This accessibility may encourage wider adoption of AI-driven security audits in the developer community.

Future developments may include improvements in the AI's ability to accurately assess vulnerability severity and impact, reducing the need for extensive human review. Additionally, the response of app developers to these AI-discovered vulnerabilities will be a key indicator of the practical impact of these tools on mobile app security.

相關指引和測驗

人工智慧代理AI 倫理人工智慧模型解釋測試你所知道的—嘗試免費的人工智慧測驗在我們的詞彙表中尋找人工智慧術語關注AI監管追蹤器
覺得有用嗎?