> ## Content Index > Fetch the complete content index at: https://www.notatechguy.com/llms.txt > Use this file to discover other available public pages before exploring further. # Tencent AI-Infra-Guard: 5,200-star AI red team tool trends - URL: https://www.notatechguy.com/tencent-ai-infra-guard-5-200-star-ai-red-team-tool-trends/ - Published: 2026-08-22T11:13:35.000Z - Updated: 2026-08-22T11:13:35.000Z - Description: Tencent Zhuque Lab's open-source tool scans AI agents, MCP servers and LLM jailbreaks. It has 5,200+ stars and a README that explicitly asks for them. - Author: Marcello Babbili - Tags: Technology & AI, AI Models, AI Agents Tencent's AI-Infra-Guard, an open-source red teaming platform that scans AI agents, MCP servers and LLM jailbreaks, landed on GitHub's daily trending chart this week with 5,206 stars [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com). The Python tool from Tencent Zhuque Lab claims to cover the full AI security stack, from agent frameworks to the Model Context Protocol ecosystem [P³](https://arxiv.org/html/2606.31227v1?ref=notatechguy.com). But the README has a line that makes the star count harder to read at face value. **My read:** This is the first AI security tool I have seen that attempts the full stack in one package: agents, skills, MCP servers, infrastructure and jailbreaks. Most competitors handle one or two layers. I do not buy the 2,000-plus CVE rules claim yet, because no independent assessment is cited anywhere in the evidence, and the README explicitly asks users to star the repo, which can inflate trending metrics. The arXiv technical report [P³](https://arxiv.org/html/2606.31227v1?ref=notatechguy.com) lends more credibility than a README alone, but I would want third-party validation before trusting this in a production environment. ## What the tool actually claims to do AI-Infra-Guard runs five scan types, according to its maintainers [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com). Agent Scan checks AI agent frameworks for vulnerabilities. Skills Scan inspects agent skills, including compiled Python bytecode (.pyc files) that could hide malicious code. MCP scan targets the Model Context Protocol, the open standard that lets AI models connect to external tools and data sources. AI Infra scan covers the underlying serving infrastructure. And LLM jailbreak evaluation tests whether a model can be prompted to bypass its safety guardrails. The tool deploys via Docker Compose and exposes a web interface at localhost:8088 [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com). It can also be called from OpenClaw chat through an aig-scanner skill [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com). The project is licensed under Apache 2.0 [P²](https://github.com/Tencent/AI-Infra-Guard/?ref=notatechguy.com), and its latest tracked version, 4.5.2, added bytecode bypass detection for skills, tool whitelisting for MCP servers in dynamic mode, and expanded its vulnerability library to more than 2,000 CVE rules [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com). The arXiv technical report from Zhuque Lab frames the problem plainly: the rapid growth of open-source AI infrastructure, from model serving engines to agent platforms to the MCP ecosystem, has outpaced the security tooling available to defend it [P³](https://arxiv.org/html/2606.31227v1?ref=notatechguy.com). That gap is real. The demand for AI security tooling is rising across the board. ## Why the star count needs an asterisk The README directly asks users to star the repository, telling readers that more stars help the project reach a wider audience and attract developers, and that each star matters to the team [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com). The source material itself notes that the README is the maintainers' own description, not an independent review [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com). That matters because trending on GitHub is driven partly by star velocity. A repo that explicitly solicits stars can climb the daily chart faster than one that does not, even if the underlying code quality is identical. The deep research snapshots show the star count moving from 4,483 [P⁴](https://github.com/tencent/ai-infra-guard?ref=notatechguy.com) to 5,206 on the trending day [S¹](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com) to 5,348 in a later snapshot [P²](https://github.com/Tencent/AI-Infra-Guard/?ref=notatechguy.com), a gain of roughly 865 stars across those snapshots. Some of that growth is organic interest in AI security. Some may be driven by the explicit ask. For comparison, Snyk's agent-scan, a competing security scanner for AI agents and MCP servers, has 2,937 stars [P⁵](https://github.com/snyk/agent-scan?ref=notatechguy.com). AI-Infra-Guard has nearly double that, but Snyk's tool does not solicit stars in its README. ![GitHub stars: AI security scanners](https://storage.ghost.io/c/6e/89/6e896869-22ef-4281-a213-b4c462c17cff/content/images/2026/08/chart_162fa31a17ce1c531273.png) ## What to do about it Consider a fintech startup building an AI assistant that connects to a customer database via MCP. Before shipping, the engineering team needs to know three things: can the agent be jailbroken into running unauthorised queries, can a malicious MCP tool execute remote code on the server, and does the infrastructure itself have known CVEs. AI-Infra-Guard claims to check all three in one Docker container. That breadth is the appeal. Running separate tools for agent scanning, MCP auditing and infrastructure CVE checks is slow and produces fragmented results. A single platform that covers the stack could save a security team days per release cycle, if the detection rules actually work. The practical step this week: clone the repo, deploy it via Docker Compose against a non-production agent setup, and run the MCP scan against a test MCP server with a deliberately vulnerable tool. If the scanner catches it, the tool has earned a closer look. If it does not, the 2,000-plus CVE rules claim is just a number on a README. ## What we don't know yet No independent security researcher has published an assessment of AI-Infra-Guard. The 2,000-plus CVE rules are not validated by any third party. The changelog dates, including a reference to 2026-08-17, may be templated or incorrectly dated, so the claimed release cadence needs checking. And the star-solicitation in the README means the trending rank reflects a mix of organic interest and active campaigning, not pure community adoption. The next signal: the next tagged release on the AI-Infra-Guard GitHub repository. We will check the release notes against the 2,000-plus CVE rule count and the MCP RCE prevention claim. Subscribe and we will send the check the day it ships. --- *Sources: [S1 — Tencent/AI-Infra-Guard: A full-stack AI Red Teaming platform securing ](https://github.com/Tencent/AI-Infra-Guard?ref=notatechguy.com) · [P2 — Tencent/AI-Infra-Guard](https://github.com/Tencent/AI-Infra-Guard/?ref=notatechguy.com) · [P3 — AI-Infra-Guard Technical Report](https://arxiv.org/html/2606.31227v1?ref=notatechguy.com) · [P4 — Tencent/AI-Infra-Guard](https://github.com/tencent/ai-infra-guard?ref=notatechguy.com) · [P5 — snyk/agent-scan](https://github.com/snyk/agent-scan?ref=notatechguy.com)* ## More from Not A Tech Guy - [Neurosymbolic world model transfers tasks without retraining](https://www.notatechguy.com/neurosymbolic-world-model-transfers-tasks-without-retraining/) - [Claude Code v2.1.239 adds cost tracking, hits 142k stars](https://www.notatechguy.com/claude-code-v2-1-239-adds-cost-tracking-hits-142k-stars/) - [FedLNS catches rogue clients in federated LLM training](https://www.notatechguy.com/fedlns-catches-rogue-clients-in-federated-llm-training/) --- *Generated from an audited evidence pack with primary-source research. Social-media items are discussion signals, not verified facts. Nothing here is financial, legal or medical advice.*