Artificial-Intelligence-Cyber-Shield - A public knowledge repository for Artificial Intel

项目说明

Artificial Intelligence Cyber Shield

Artificial Intelligence Cyber Shield is a comprehensi knowledge base dedicated to AI Security, AI Safety, AI Go, AI Assurance, AI Agent Security, AI Red Teaming, and Cyber Defense in the age of intelligent systems.

The repository brings together leading international frameworks, standards, threat models, security controls, red team methodologies, incident repositories, go guidance, professional certifications, organizational certifications, assurance programs, and engineering best practices into a single reference for researchers, security professionals, policymakers, architects, CISOs, and national security practitioners.

The project is part of the broader GAIA Cybersecurity decision-support concept and supports structured, e adoption of trustworthy and secure AI.

Purpose

This repository helps security leaders, researchers, engineers, architects, red teams, blue teams, go teams, and national-security practitioners organize AI security, AI safety, and AI go knowledge in a structured and defensible way.

The main goal is to connect AI go, AI safety, AI threat modeling, red teaming, assurance, model security, agent security, AI integrity, professional de, organizational certification, and cyber defense into one practical knowledge base.

Scope

| Area | Focus |
| --- | --- |
| AI security go | Policies, controls, accountability, risk ownership, and o |
| AI safety | Alignment, robustness, reliability, e, human o, and pre of unintended harm |
| AI risk management | AI RMF, assurance methods, lifecycle controls, and risk documentation |
| AI project failure and success | Root causes, anti-patterns, production readiness, lifecycle gates, value realization, and stop criteria |
| AI red teaming | Testing AI systems for misuse, abuse, prompt injection, jailbreaks, data leakage, unsafe behavior, e awareness, sandbagging, and alignment faking |
| AI threat modeling | Structured analysis of threats against models, agents, data pipelines, APIs, plugins, and integrations |
| AI agent security | Tool use, autonomy, memory, identity, permissions, delegation, containment, external hostile agents, and high-risk capability control |
| AI integrity | Protection against model backdoors, secret or unauthorized behavior changes, poisoned artifacts, prompt/policy tampering, and harness sub |
| Secure AI engineering | Secure design, secure deployment, guardrails, monitoring, e, and logging |
| System hardening against AI threats | Mandatory access control, sandboxing, execution control, isolation, Zero Trust networking, and controlled egress |
| AI for cyber defense | Detection, triage, prioritization, SOC assistance, defensi, detection-in-depth, and differential access |
| Cybersecurity for AI | Protection of AI infrastructure, training data, model artifacts, inference endpoints, internal frontier models, and supply chains |
| AI assurance | E, , independent review, transparency, traceability, safety cases, and e controls |
| AI supply chain security | Model pro, datasets, dependencies, SBOM/AIBOM, integrity , and model distribution |
| AI incident management | AI incident repositories, lessons learned, disclosure practices, autonomous-agent containment failures, and response processes |
| AI threat landscape | Covert AI-enabled operations, HACCAs, autonomous cyber agents, loss-of-control scenarios, and persistent machine-speed ad |
| Professional certifications | AI security, AI safety, go, risk, audit, assurance, and ad testing credentials |
| Organizational certifications and assurance | ISO/IEC 42001, CSA STAR for AI, accreditation, registry assurance, and supporting organizational certifications |

Core framework pages

| Page | Focus |
| --- | --- |
| AI Security Frameworks | NIST, OWASP, MITRE ATLAS, CSA, Google SAIF, Pillar Security SAIL, ISO, ENISA, ETSI, CISA/NCSC, OECD, and EU AI Act references |
| AI Incident Response Standards and Frameworks | Direct AI-specific incident reporting, incident information-sharing, and regulatory reporting frameworks from ETSI, CISA, OECD, and the EU AI Act |
| AI Safety vs AI Security | Comparison of AI safety and AI security, including objectives, risks, controls, and complementary roles |
| AI Red Team Frameworks | AI red teaming methods, benchmarks, tooling, frontier-model e resources, e awareness, sandbagging, and alignment-faking tests |
| AI Threat Modeling Frameworks | AI STRIDE, MITRE ATLAS, OWASP, PASTA, LINDDUN, and AI-specific threat modeling layers |
| AI Agent Security | Agent identity, tool use, memory, RAG, autonomy, MCP-style integrations, containment, hostile external agents, HACCAs, and Agent Rule of One |
| AI Integrity | Backdoors, secret loyalties, model and artifact tampering, prompt/policy integrity, harness integrity, beha , and rollback |
| AI for Cyber Defense and Agentic SOC Frameworks | ATHF, ADEF, IAPS detection-in-depth, agentic SOC security, and differential access to cyber-capable models |
| CUSTODY Autonomous Agent Containment Framework | Agent authority, reach, runtime containment, egress, stop controls, and July 2026 autonomous-agent incident mapping |
| System Hardening Against AI and Agentic AI Threats | SELinux, AppArmor, seccomp, Landlock, Windows controls, container isolation, controlled egress, and human approval |
| AI Go and Assurance | NIST AI RMF, ISO/IEC 42001, EU AI Act, GPAI Code of Practice, ETSI, CSA, frontier safety frameworks, and assurance e |
| AI Go, Risk, Security, and Assurance Frameworks | Standards and policy catalog including current U.S. EO 14409, historical status of EO 14110, and internal frontier-model risk reporting |
| AI Threat Landscape | Covert AI-enabled attacks, HACCAs, autonomous cyber-agent containment, and the July 2026 OpenAI/Hugging Face incident |
| Why AI Projects Fail and How They Can Succeed | RAND root causes, organizational anti-patterns, cybersecurity implications, lifecycle stage gates, metrics, and executive stop criteria |
| Professional AI Security and AI Safety Certifications | Comparison of ISACA, Cloud Security Alliance, GIAC, OffSec, IAPP, uni programs, and ISO/IEC 42001 professional credentials |
| Organizational AI Certifications and Assurance Programs | ISO/IEC 42001 certification, CSA STAR for AI levels, AICM, AI-CAIQ, accreditation, and assurance |

Additional knowledge areas

AI standards and guidance

Co includes standards, guidance, and research from:

| Organization | Relevance |
| --- | --- |
| NIST | AI RMF, Generative AI Profile, cybersecurity and risk management guidance |
| ISO/IEC | AI management systems, AI risk management, security, privacy, and go standards |
| ETSI | AI cybersecurity, Securing AI, testing, safety, and trustworthiness guidance |
| ENISA | AI cybersecurity, threat landscape, and EU cybersecurity guidance |
| OWASP | LLM application security, agentic AI security, and application threat guidance |
| MITRE | ATLAS ad threat knowledge base and AI attack techniques |
| Cloud Security Alliance | AI Controls Matrix, AI safety and security guidance, enterprise control mapping |
| OECD | AI Principles and AI incident monitoring |
| European Union | EU AI Act, GPAI Code of Practice, and digital regulation context |
| CISA and NCSC | Secure AI deployment, secure de, and operational resilience guidance |
| Pillar Security | SAIL Secure AI Lifecycle Framework for lifecycle-oriented AI and agentic AI security |
| Institute for AI Policy and Strategy (IAPS) | Autonomous cyber agents, HACCAs, detection-in-depth, AI integrity, e awareness, differential access, internal-model risk reporting, and frontier-security policy research |
| Google, Microsoft, Anthropic, and OpenAI | Frontier AI safety, red teaming, secure AI de, and preparedness practices |

AI security frameworks

Examples include:

1. NIST AI RMF.
2. NIST AI 600-1 Generative AI Profile.
3. Google Secure AI Framework.
4. MITRE ATLAS.
5. OWASP Top 10 for LLM Applications.
6. OWASP Agentic AI guidance.
7. Cloud Security Alliance AI Controls Matrix.
8. Pillar Security SAIL Secure AI Lifecycle Framework.
9. ETSI AI security standards and guidance.
10. ISO/IEC 42001.
11. ISO/IEC 23894.
12. CUSTODY autonomous-agent containment.
13. IAPS offensi agent detection-in-depth.

AI safety

Topics include:

1. Alignment.
2. Robustness.
3. Reliability.
4. Human o.
5. Safe deployment.
6. Interpretability.
7. Frontier model safety.
8. E methodologies.
9. Responsible scaling.
10. Safety cases and assurance e.
11. E awareness.
12. Sandbagging and alignment faking.
13. Post-deployment monitoring for e behavior differences.

AI threat modeling

Includes:

1. AI STRIDE.
2. MITRE ATLAS.
3. PASTA.
4. LINDDUN.
5. Attack trees.
6. Kill chain adaptations.
7. Agentic AI threat models.
8. Trust boundary analysis.
9. Abuse case modeling.
10. Data, model, agent, tool, and infrastructure layers.
11. HACCA campaign sustainability and shutdown-a behaviors.
12. Autonomous-agent containment and target-scope escape.

AI red teaming

Co includes:

1. Microsoft AI Red Team.
2. PyRIT.
3. garak.
4. OpenAI Preparedness e.
5. Anthropic e.
6. Google Frontier Safety Framework.
7. NIST GenAI testing guidance.
8. Prompt injection testing.
9. Jailbreak testing.
10. Misuse and abuse testing.
11. Hidden and held-out e.
12. E and sandbagging tests.
13. Long-horizon containment and target-boundary tests for cyber-capable agents.

System hardening against AI threats

The System Hardening Against AI and Agentic AI Threats page covers:

1. Threat-to-control mapping for prompt injection, generated code, credential theft, data exfiltration, pri escalation, and lateral mo.
2. SELinux, AppArmor, seccomp, Landlock, systemd sandboxing, namespaces, and cgroups.
3. Windows Application Control, AppContainer, PowerShell restrictions, and least-pri service identities.
4. Container and Kubernetes security contexts, default-deny network policy, and stronger isolation for untrusted code.
5. Controlled egress, tool gateways, scoped credentials, human approval, monitoring, response, and red-team .
6. Independent target allowlists and network containment for cyber-capability e.

AI integrity

The AI Integrity page covers:

1. Training-data poisoning and malicious fine-tuning.
2. Model backdoors and hidden triggers.
3. Secret loyalties or covert beha objectives.
4. Model-weight and artifact tampering.
5. System-prompt and policy integrity.
6. Agent harness, tool, identity, memory, and RAG integrity.
7. E integrity and sabotage.
8. Supply-chain pro and signed artifacts.
9. Runtime configuration integrity.
10. Known-good rollback and recovery.

AI project failure and success

The Why AI Projects Fail and How They Can Succeed page covers:

1. The RAND study by Ryseff, De Bruhl, and Newberry and its fi root causes.
2. The distinction between model performance, operational value, and project success.
3. Leadership, data, technology-first, infrastructure, and technological-maturity anti-patterns.
4. The role of cybersecurity, privacy, safety, go, and assurance in production readiness.
5. Stage gates from problem definition and data readiness through controlled deployment, operation, and retirement.
6. Mapping to NIST AI RMF GOVERN, MAP, MEASURE, and MANAGE functions.
7. Balanced success metrics, executi questions, early-warning indicators, and stop criteria.

Professional certifications

The Professional AI Security and AI Safety Certifications page compares credentials and professional programs for:

1. Enterprise AI security leadership.
2. AI safety and cloud security.
3. Defensive AI application security.
4. AI red teaming and ad testing.
5. AI go and responsible AI.
6. AI risk management.
7. AI audit and assurance.
8. ISO/IEC 42001 implementation and auditing.
9. AI-enabled defensive and offensi operations.
10. Applied uni AI security education.

Organizational certifications and assurance

The Organizational AI Certifications and Assurance Programs page covers:

1. ISO/IEC 42001 Artificial Intelligence Management System certification.
2. CSA STAR for AI Level 1, , and Level 2 assurance pathways.
3. CSA AI Controls Matrix and AI-CAIQ e.
4. ISO/IEC 42006 requirements for AIMS audit and certification bodies.
5. Accreditation, certification-scope, registry, and certificate- considerations.
6. Supporting certifications such as ISO/IEC 27001, ISO/IEC 27701, ISO 9001, CSA STAR for cloud services, and SOC 2 reporting.
7. A staged organizational roadmap from self-assessment to independent certification and ad registry assurance.

AI incident repositories

Including:

1. AI Incident Database.
2. OECD AI Incidents Monitor.
3. Public AI safety and security incident repositories.
4. Case studies on model failure, abuse, data leakage, misuse, and containment failure.
5. OWASP Agentic AI Security Incidents & Exploits.
6. IAPS analysis of the July 2026 OpenAI/Hugging Face autonomous-agent incident.

Repository roadmap

1. Expand AI security and go frameworks.
2. Expand AI safety frameworks and safety assurance guidance.
3. Expand AI red teaming frameworks and methodologies, including e testing.
4. Expand AI threat modeling frameworks, including AI STRIDE, agentic AI threat models, and HACCA scenarios.
5. Expand agent security principles, control patterns, detection-in-depth, and maturity models.
6. Maintain system-hardening guidance for AI and agentic workloads across Linux, Windows, containers, and cloud en.
7. Maintain AI integrity guidance co models, prompts, policies, tools, memory, RAG, and runtime configuration.
8. Expand AI project failure, success, production-readiness, and benefit-realization guidance.
9. Add and expand AI supply chain security guidance.
10. Add AI incident repositories and case studies, including autonomous-agent containment failures.
11. Expand ETSI, ISO/IEC, ENISA, OWASP, MITRE, CSA, Pillar Security, NIST, and IAPS mapping.
12. Maintain and expand the professional certification comparison.
13. Maintain and expand organizational certification, accreditation, and assurance-program guidance.
14. Maintain current legal and policy status, including re, replacements, and new executi.
15. Add references in APA 7 style.
16. Add diagrams and decision-support tables.
17. Publish the content as a readable GitHub Pages knowledge base.

Suggested structure

项目信息

开发语言：未知

Star 数：4

技术要点

该项目属于防护思路方向的开源资源，代码结构清晰，适合安全研究人员学习参考。项目围绕A public knowledge repository for Artificial Intelligence Cyber Shield architectures, security frameworks, defensive technologies, references, and analytical notes.展开，提供了可运行的实现，对理解相关安全机制、工具原理与攻防技术具有较高的研究价值。

从技术实现角度看，此类项目的核心在于对底层原理的深入理解与工程化落地。开发者在研究过程中可以结合官方文档与社区讨论，梳理项目的初始化流程、核心模块职责与数据流转，逐步掌握其中的技术细节与设计思路。

使用与学习建议

1. 下载源码压缩包，解压到本地环境。

2. 根据项目类型准备对应的运行环境与依赖。

3. 按照 README 中的步骤完成配置与部署。

4. 在隔离的测试环境中运行验证，避免影响生产系统。

免责声明

本资源来源于开源社区，仅供安全研究、技术学习与合法授权范围内的攻防演练使用，请勿用于任何非法或侵害他人权益的用途。使用本项目产生的一切后果由使用者自行承担。