AI-Infra-Guard

No entries match these filters. Clear filters

1. v4.6.8 Oct 29, 2026 · issue 124

   AI-Infra-Guard adds a Runcrate provider preset to Agent-Scan and packages each scanner for independent pip install⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.8 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.8

   • Install any scanner on its own without pulling the full tool by packaging each scanner for an independent pip install.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Agent-Scan now supports Runcrate as a provider preset, expanding the range of agent runtimes it can target.
   • Install any scanner on its own without pulling the full tool by packaging each scanner for an independent pip install.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   1 more in Everything

2. v4.6.7 Oct 22, 2026 · issue 116

   Agent-Scan gains AG-UI (CopilotKit) protocol stream support, a cheaper inference provider preset, and daily AI CVE advisories plus astron-agent fingerprinting.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.7 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.7

   • Adds astron-agent fingerprint to the data layer for AI infrastructure detection.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds a Cheaper Inference provider preset to Agent-Scan for cost-sensitive scan configurations.
   • Agent-Scan now forwards the prompt and selected skills to the scanner, giving operators direct control over scan inputs.
   • Supports AG-UI (CopilotKit) protocol streams in Agent-Scan, enabling scanning of CopilotKit-backed agents.
   • Adds astron-agent fingerprint to the data layer for AI infrastructure detection.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Supports AG-UI (CopilotKit) protocol streams in Agent-Scan, enabling scanning of CopilotKit-backed agents.

   2 more in Everything

3. v4.6.6 Oct 15, 2026 · issue 109

   CLI gains JSON scan output; Dashboard model dialog gains extra_headers/extra_body editing; Eval adds BadPhoneAgent bilingual benchmark.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.6 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.6

   • Adds JSON scan output via the CLI, giving practitioners machine-readable results for pipeline integration.
   • Adds extra_headers / extra_body editing to the Dashboard model dialog, letting operators inject custom HTTP headers and body fields per model provider.
   • Adds BadPhoneAgent bilingual benchmark to Eval, extending jailbreak evaluation coverage with a new dual-language phone-agent attack scenario.

   3 details in Everything

   • New detection Memory/RAG poisoning detection skill
   • New integration Atlas Cloud provider preset for Agent-Scan
   • New integration API Route provider preset for Agent-Scan
4. v4.6.4 Oct 1, 2026 · issue 091

   Adds pikit, a composable prompt-injectionprompt-injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. research toolkit, under Research/pikit⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.4 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.4

5. v4.6.3 Sep 24, 2026 · issue 084

   DeepTeam now flushes finished test cases to disk mid-run; API Checker gets more reliable model fingerprinting via sample refilling.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.3 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.3

   • Adds mid-run disk persistence for DeepTeam: finished test cases are written to disk during a run rather than held only in memory.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Improves API Checker model identification by refilling failed fingerprint samples for more reliable detection.
   • Adds mid-run disk persistence for DeepTeam: finished test cases are written to disk during a run rather than held only in memory.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   1 more in Everything

6. v4.6.2 Sep 17, 2026 · issue 077

   AI-Infra-Guard v4.6.2 adds FORGE-Bench loss-of-control benchmark and RogueHandoff-20 benchmark for autonomous agent red teaming.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.2 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.2

   • Adds forge_bench (FORGE-Bench) — a deterministic Loss-of-Control benchmark for evaluating autonomous agents, with a built-in OpenAI-compatible client and bilingual quickstart.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds RogueHandoff-20 benchmark under the Research module for autonomous agent handoff security evaluation.
   • Updates AIG rules dataset (2026-09-11 ruleset).
   • Adds forge_bench (FORGE-Bench) — a deterministic Loss-of-Control benchmark for evaluating autonomous agents, with a built-in OpenAI-compatible client and bilingual quickstart.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Adds RogueHandoff-20 benchmark under the Research module for autonomous agent handoff security evaluation.

   1 more in Everything

7. v4.6.1 Sep 10, 2026 · issue 070

   AI-Infra-Guard v4.6.1 adds Gemini, Gemma, and GLM fingerprints to the API Checker.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.1 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.1

   • Adds Gemini, Gemma, and GLM fingerprints to the API Checker for model identification.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠
   • Updates the GLM-5.3-Flash fingerprint baseline in the API Checker.
   • Adds Gemini, Gemma, and GLM fingerprints to the API Checker for model identification.⁠︀︄︇︀︅︉︀︅︀︀︃︀︀︅︅︀︅︊︀︄︆︀︃︄⁠

   1 more in Everything

8. v4.6.0 Aug 26, 2026 · issue 058

   AI-Infra-Guard v4.6.0 adds an API security audit module, LLMLLMA large language model, a neural network trained on large amounts of text to recognize patterns and generate language. Applications use it for tasks such as answering, summarizing and writing code. API poisoning detection, and a v5.0.0 mutation engine for agent red-teaming.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.6.0 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.6.0

   • Adds detection for LLM API poisoning attacks.
   • Merges workflow-attack into mutation-attack in the aig-agent-redteam v5.0.0 mutation engine refactor.
   • Adds strict YAML validation enforcing required id and severity fields in vulnerability rules.
   • Includes DeepSeek Harness prompt injectionprompt injectionAn attack where text the model reads, such as a web page, file or tool result, carries instructions that override the user's. It is the main security risk for agents with tools. assessment research.
   • Updates AIG vulnerability rules with two new rule-set snapshots dated 2026-08-07 and 2026-08-14.

   5 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !The workflow-attack subcommand/mode in aig-agent-redteam is merged into mutation-attack; any automation invoking workflow-attack will need to be updated.
9. v4.5.2 Aug 17, 2026 · issue 048

   AI-Infra-Guard v4.5.2 adds SkillJack research, GET-only fingerprints for Qdrant/Chroma/Weaviate, and a v5.0.0 mutation engine refactor merging workflow-attack into mutation-attack.

   └──▷ GET THIS VERSION

   copy

   $ git clone --branch v4.5.2 https://github.com/Tencent/AI-Infra-Guard.git
   # already have the repo? check out this version:
   $ git checkout v4.5.2

   • Adds GET-only fingerprints for Qdrant, Chroma, and Weaviate to the fingerprint library, enabling passive identification of these vector databasesvector databasesA database built to store embeddings and find the nearest ones to a query quickly. It is the retrieval half of most RAG systems..
   • Introduces the SkillJack research project, expanding the platform's AI security research surface.
   • Upgrades the aig-agent-redteam skill to v5.0.0 with a refactored mutation engine that merges workflow-attack into mutation-attack for consolidated adversarial testing.
   • Adds new AIG detection rules dated 2026-07-31.

   4 details in Everything

   └──▷ BREAKING ON UPGRADE

   • !The aig-agent-redteam skill's workflow-attack has been merged into mutation-attack; any automation or config referencing workflow-attack as a separate attack mode will need to be updated.
10. v4.5.1 Jul 30, 2026 · issue 030

    AI-Infra-Guard v4.5.1 adds multi-turn jailbreak attacks, 6 new OWASPOWASPA nonprofit foundation that publishes open standards, guides, and ranked vulnerability lists—most famously the OWASP Top 10—giving cyber tools a vendor-neutral benchmark to test or report against. agent-scan skills, and 4 new MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. security detection rules.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.5.1 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.5.1

    • Adds Many-Shot, PAIR, GOAT, and ActorAttack multi-turn jailbreak attack methods to PromptSecurity.
    • Adds 5 new OWASP-aligned Agent-Scan detection skills: agentic-supply-chain, cascading-failure, human-agent-trust, inter-agent-comm, and unexpected-code-execution.
    • Adds web-exfiltration-detection skill to Agent-Scan.
    • Adds 4 new MCP-Scan security detection rules covering hardcoded secrets and insecure deserializationdeserializationTurning stored or transmitted data back into objects a program can use. Unsafe handling of untrusted data can allow unexpected behavior or code execution..

    4 details in Everything

11. v4.5.0 Jul 27, 2026 · issue 027

    AI-Infra-Guard v4.5.0 adds standalone CLI modules for Agent/MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface./Skill scanning, new AI agent detection skills, SARIFSARIFStatic Analysis Results Interchange Format, a standard for sharing code-analysis findings, including locations and rule details. It lets scanners report results to a common review interface. output, and open-source frontend.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.5.0 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.5.0

    • Modularizes Agent-Scan as a standalone CLI with AIG integration support, adding 4 new AI agent security detection skills.
    • Modularizes MCP-Scan with dual-mode support (CLI + AIG Web) and adds a standalone mcp-scan-lite module.
    • Adds 2 new MCP security detection rules plus ATR-derived MCP detection rules covering additional attack surfacesattack surfacesThe exposed systems, interfaces and access points an attacker could try to use. It includes internet-facing services as well as identities, software and other reachable resources..
    • Adds Agent Skill security auditing (Skill-Scan) repackaged as the standalone PyPI package aig-skill-scan.
    • Adds SARIF 2.1.0SARIF 2.1.0A JSON-based file format, standardized by OASIS, for expressing static analysis results so that cyber tools from different vendors can share findings through a single, machine-readable schema. output format for Skill-Scan results alongside single-stage optimization.

    +4 more show less

    • Adds Stage 2 Code Audit Markdown report output for Skill-Scan (replacing XML).
    • Adds agentic-tool-misuse evaluation dataset for assessing agentic tool misuse scenarios.
    • Open-sources the frontend code with open-source environment configuration.
    • Adds new AI component fingerprints and AIG vulnerability rules from June–July 2026 rule updates.

    9 details in Everything

12. v4.1.15 Jun 25, 2026 · release notes

    Adds mcp_scan and ai_infra_scan model token flexibility; new MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. threat detections

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.15 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.15

    • mcp_scan and ai_infra_scan commands now allow omitting the model token argument to fall back to the system default model instead of failing.
    • New MCP threat detection rules added for tool poisoning, credential exfiltration, and command injectioncommand injectionAn attack that makes an application execute operating-system commands supplied through untrusted input. It often happens when input is passed unsafely to a shell..

    2 details in Everything

13. v4.1.15 Jun 25, 2026 · issue 020

    AI-Infra-Guard v4.1.15 adds optional model token in scan APIs, 6 new llama.cpp CVE rules, and 3 new MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. threat detection rules.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.15 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.15

    • Supports omitting the model token in mcp_scan and ai_infra_scan API calls, falling back to the system default model.
    • Adds 6 new llama.cpp CVE detection rules to the llama-cpp vulnerability pack.
    • Adds 3 new MCP threat detection rules covering tool poisoning, credential exfiltration, and command injectioncommand injectionAn attack that makes an application execute operating-system commands supplied through untrusted input. It often happens when input is passed unsafely to a shell..

    3 details in Everything

14. v4.1.14 Jun 18, 2026 · release notes

    AI-Infra-Guard v4.1.14 adds aig-agent-redteam skill, 9 new jailbreak methods, and eval datasets

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.14 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.14

    • Adds aig-agent-redteam skill for Agent security assessment
    • Adds 9 single-turn jailbreak attack methods: PrefillAttack, ICA, PastTense, Overload, Jailbroken, FlipAttack, DeepInception, CodeChameleon, JAM
    • Adds jailbreak evaluation datasets AdvBench, CNSafe, SafeBench

    3 details in Everything

15. v4.1.13 Jun 11, 2026 · release notes

    AI-Infra-Guard v4.1.13 adds a version check API endpoint

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.13 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.13

16. v4.1.12 Jun 8, 2026 · release notes

    v4.1.12 adds 39 new AI web fingerprints and enhances 18 existing ones for AI infrastructure detection

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.12 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.12

    • Adds support for 39 new AI web fingerprints to detect additional AI services and frameworks

    1 detail in Everything

17. v4.1.10 May 28, 2026 · release notes

    Adds --websocket support for agent providers in AI-Infra-Guard v4.1.10

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.10 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.10

    • Supports WebSocketWebSocketA protocol that upgrades an HTTP connection into a persistent, full-duplex channel, letting a server push data to a client without repeated polling. Cyber tools use it for real-time event streaming and live command output. agent providers, enabling connections to inference endpoints that communicate over the WebSocket protocol.

    1 detail in Everything

18. v4.1.9 May 21, 2026 · release notes

    AI-Infra-Guard v4.1.9 adds 20+ single-turn and 6 multi-turn prompt attack operators for red-teaming.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.9 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.9

    • Adds invisible-text, case-formatting, script-system, unicode-style, classical-cipher, classic-encoding, SystemOverride, SuperUser, LinguisticConfusion, Roleplay, PromptProbing, PromptInjection, PROMISQROUTE, PermissionEscalation, Multilingual, MathProblem, InputBypass, ICRTJailbreak, GrayBox, GoalRedirection, EquaCode, and ContextPoisoning as single-turn prompt attack operators.
    • Adds TreeJailbreaking, SequentialJailbreaking, LinearJailbreaking, CrescendoJailbreaking, BestofN, and BadLikertJudge as multi-turn prompt attack operators.
    • Registers newly added attack operators in the documentation.

    3 details in Everything

19. v4.1.6 Apr 23, 2026 · release notes

    v4.1.6 adds detection for exposed AI agent configs (claude_desktop_config.json, mcp.json), manual dataset updates, and merges the update-status API into update-data.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.6 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.6

    • Adds update-data endpoint merging previous update-status functionality
    • Supports manual updates to latest jailbreak datasets, fingerprints, and vulnerability databases
    • Detects exposed AI agent config files like claude_desktop_config.json and mcp.json
    • Extends .env key patterns for broader secret coverage

    4 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The /update-status endpoint is removed; use update-data instead.
20. v4.1.4 Apr 17, 2026 · release notes

    v4.1.4 adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. red team scan, data sync APIs, and self-signed TLS support for model endpoints

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.4 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.4

    • Add --agent-config-file CLI flag to pass inline YAML agent scan config (requires aig-scanner v1.0.3)
    • Add InsecureSkipVerify option to bypass TLS verification for HTTPSHTTPSHTTP carried over an encrypted TLS connection. It protects traffic in transit and lets a browser verify the server's certificate. connections with self-signed/private CA certificates
    • Add multi-turn red team attack module supporting TAP and Crescendo strategies in MCP Scan
    • Add POST /api/v1/system/update-data and GET /api/v1/system/update-status endpoints for auto-syncing the data/ directory
    • Support inline agent_config in Agent Scan API with optional verify flag on save

    5 details in Everything

21. v4.1.2 Apr 3, 2026 · release notes

    AI-Infra-Guard v4.1.2 adds support for stopping running tasks via task control

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.2 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.2

    • Added support for stopping running tasks, allowing users to actively terminate scans in progress

    1 detail in Everything

22. v4.1.1 Mar 25, 2026 · release notes

    AI-Infra-Guard v4.1.1 adds FOFA API fingerprint matching syntax and new AIG vulnerability rules.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1.1 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1.1

    • Adds correct new-api fingerprint matcher syntax (FOFA 100%) to enable precise fingerprinting via FOFA
    • Integrates AIG rules batch from 2026-03-25 to expand AI component vulnerability detection

    2 details in Everything

23. v4.1 Mar 23, 2026 · release notes

    AI-Infra-Guard v4.1 adds OpenClaw coverage, EdgeOne ClawScan skill, and task API enhancements

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.1 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.1

    • Adds port 18789 to the default AI infrastructure scan port list
    • AIG Rules (2026-03-20 batch) for AI component vulnerability detection
    • New OpenClaw vulnerability database with 281 CVE/GHSA entries
    • Adds EdgeOne-based ClawScan security scanning Skill to agent capabilities
    • Task API now supports enhanced Agent Scan and lifecycle management

    5 details in Everything

24. v4.0 Mar 10, 2026 · release notes

    AI-Infra-Guard v4.0 introduces the Agent-Scan framework, ClawScan integration, and new AI component fingerprints.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v4.0 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v4.0

    • EdgeOne ClawScan provides a dedicated security portal for OpenClaw via [https://matrix.tencent.com/clawscan/](https://matrix.tencent.com/clawscan/openclaw).
    • New Agent-Scan framework introduces a multi-agent architecture with specialized agents (SSRFSSRFServer-side request forgery, an attack that tricks a server into making requests chosen by an attacker. Those requests can reach internal services the attacker cannot access directly., config-scanner, vulnerability detector, etc.) for autonomous agent security scanning.
    • Added fingerprints for llama.cpp, HuggingFace TGI, NVIDIA NIM, and LocalAI in component detection.
    • Agent-Scan simulates interactions using actions: bash, file read/write, edit, grep, glob, ls, batch, thinking, todo, task, skill, MCP tool, dialogue, and finish.

    4 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !Removed the idSuffix field from configuration.
25. v3.6.2 Feb 9, 2026 · release notes

    Adds n8n fingerprint for AI infrastructure scanning

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v3.6.2 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v3.6.2

    • Added n8n fingerprint to detect n8n-based automation workflows in scanned environments

    1 detail in Everything

26. v3.6.0 Jan 17, 2026 · release notes

    AI-Infra-Guard v3.6.0 adds inner API auth controller and SYS_ADMIN capability for Chrome sandbox.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v3.6.0 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v3.6.0

    • Adds inner API auth controller to handle authentication for internal API endpoints
    • Adds SYS_ADMIN capability for Chrome sandbox and database indexes for performance enhancement

    2 details in Everything

27. v3.5.0 Dec 26, 2025 · release notes

    AI-Infra-Guard v3.5.0 adds dynamic analysis framework, MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. remote tool invocation, and new detection types

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v3.5.0 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v3.5.0

    • Adds Dynamic Analysis Framework with specialized agents for malicious behavior and vulnerability testing
    • Adds tool poisoning detection and rug pull detection capabilities
    • Adds support for new evaluation datasets including copyright-violation, misinformation, privacy-leakage, unethical-behavior, violent, and non-violent-illegal-activity
    • Adds mcp_tool for remote MCP server tool invocation
    • Adds write_file tool for file writing operations

    +4 more show less

    • Adds version router endpoint /version via the Version API
    • Adds prompt_manager utility for prompt template management
    • Adds custom MCP Header Support for authentication and protocol configuration
    • Adds frontend integration for MCP scan header configuration in the settings panel

    9 details in Everything

28. v3.5-preview Prerelease Dec 4, 2025 · release notes

    New MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface.-Scan framework, headless screenshot capabilities, and advanced encoding/obfuscation attack methods in AI-Infra-Guard v3.5-preview

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v3.5-preview https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v3.5-preview

    • Adds MCP-Scan framework for Model Context ProtocolModel Context ProtocolModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. security scanning with autonomous agent-based code audit and vulnerability review
    • Adds support for 12+ new encoding/obfuscation attack methods including A1Z26, AffineCipher, AsciiSmuggling, Aurebesh, Caesar, Leetspeak, MirrorText, Ogham, Vaporwave, Zalgo, Stego, and StrataSword suite
    • Adds Chromium-based headless screenshot functionality for web scanning

    3 details in Everything

29. v3.4.3 Oct 27, 2025 · release notes

    AI-Infra-Guard v3.4.3 adds Model Invocation Base Class, expanded API docs, and new evaluation datasets.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v3.4.3 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v3.4.3

    • Adds ModelInvocationBaseClass base class methods for model invocation to improve code reusability
    • Updates API documentation with enhanced Swagger specifications
    • Expands evaluation datasets with Cyberattack and CBRN weapons test data

    3 details in Everything

30. v3.4 Sep 18, 2025 · release notes

    AI-Infra-Guard v3.4 adds Docker ARMARMA family of RISC processor architectures designed by Arm Holdings and licensed to chip makers worldwide; tools target it to run natively on mobile, embedded, and increasingly server and desktop hardware. support, task concurrency limits, and i18n to the CLI.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v3.4 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v3.4

    • Adds Apple ARM architecture deployment support via the updated one-click Docker script
    • Adds frontend internationalization (i18n) support for multi-language text display

    2 details in Everything

31. v3.2 Aug 26, 2025 · release notes

    AI-Infra-Guard v3.2 adds model concurrency limit to MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. scans and improves report display

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v3.2 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v3.2

    • Optimizes MCP Scan Report data dimensions and narrow-screen layout

    1 detail in Everything

32. v2.5 Jun 12, 2025 · release notes

    AI-Infra-Guard v2.5 adds file size to list_dir, multi-param agent support, and MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. prompt improvements

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.5 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v2.5

    • list_dir now displays file sizes in its output
    • Agent framework refactored to support multiple parameters
    • MCP plugin prompts optimized for better interaction

    3 details in Everything

33. v2.4.1 Jun 6, 2025 · release notes

    AI-Infra-Guard v2.4.1 adds parallel processing for MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. scanning to reduce latency.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.4.1 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v2.4.1

    • Enables parallel execution for MCP (Model Context ProtocolModel Context ProtocolModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface.) scans to significantly improve scan throughput.

    1 detail in Everything

34. v2.1 Apr 29, 2025 · release notes

    AI-Infra-Guard v2.1 adds MCPMCPModel Context Protocol, a standard for connecting AI applications to external tools and data sources. A compatible server exposes capabilities an assistant can use through a common interface. history saving and WebSocketWebSocketA protocol that upgrades an HTTP connection into a persistent, full-duplex channel, letting a server push data to a client without repeated polling. Cyber tools use it for real-time event streaming and live command output. stop support

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v2.1 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v2.1

    └──▷ TRY IT
    Enable history persistence for MCP plugin sessions
    copy

    $ ai-infra-guard mcp --save-history true

    • mcp plugins now save conversation history via a dedicated parameter

    1 detail in Everything

35. v0.1 Mar 7, 2025 · release notes

    AI-Infra-Guard v0.1 initial release adds Xinference fingerprint support and updated vulnerability detection

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.1 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v0.1

36. v0.0.7 Feb 26, 2025 · release notes

    AI-Infra-Guard v0.0.7 adds DeepSeek support and rewrites -list-vul output style

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.0.7 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v0.0.7

    • Adds deepseek model support for vulnerability fingerprinting queries.
    • Rewrites the output format of the -list-vul argument to display updated fingerprint information.

    2 details in Everything

    └──▷ BREAKING ON UPGRADE

    • !The --list-fps flag has been removed; use -list-vul instead.
37. v0.0.6 Feb 21, 2025 · release notes

    AI-Infra-Guard v0.0.6 adds POST fingerprint probing, websocket server, and /show route.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.0.6 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v0.0.6

    • Adds POST request support for sending fingerprints (指纹发包)
    • Adds a WebSocketWebSocketA protocol that upgrades an HTTP connection into a persistent, full-duplex channel, letting a server push data to a client without repeated polling. Cyber tools use it for real-time event streaming and live command output. server to the tool's webserver
    • Adds /show route to the webserver

    3 details in Everything

38. v0.0.4 Feb 13, 2025 · release notes

    AI-Infra-Guard v0.0.4 adds -localscan flag for one-click local app scanning and table output format

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.0.4 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v0.0.4

    └──▷ TRY IT
    Scan a local application directory in one step without configuring remote sources.
    copy

    $ AI-Infra-Guard --localscan

    • Adds --localscan (aliased as -localscan) to perform a one-click scan of local applications
    • Introduces a new table output format option for scan results

    2 details in Everything

39. v0.0.3 Feb 7, 2025 · release notes

    v0.0.3 adds security advice to output and ClickHouse fingerprint matching.

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.0.3 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v0.0.3

    • Output now includes security suggestions (输出加入安全建议) to guide remediation
    • Adds clickhouse fingerprint support with case-insensitive fuzzy matching (添加clickhouse指纹, fuzzy匹配忽略大小写)
    • Updated vulnerability fingerprints for broader detection coverage

    3 details in Everything

40. v0.0.2 Jan 22, 2025 · release notes

    AI-Infra-Guard v0.0.2 adds -check-vul flag for vulnerability checking and expands fingerprint support

    └──▷ GET THIS VERSION

    copy

    $ git clone --branch v0.0.2 https://github.com/Tencent/AI-Infra-Guard.git
    # already have the repo? check out this version:
    $ git checkout v0.0.2

    • New fingerprint added for openwebui version identification

    1 detail in Everything