Awesome Hacking with AI

August 28, 2026 · View on GitHub

Awesome Hacking with AI Logo


MIT License PRs Welcome Awesome Stars Last Updated


Awesome Hacking with AI

A curated, evidence-led guide to AI for security and security for AI systems. It brings together authorized security testing, application security, AI agents, model adaptation, datasets, benchmarks, MCP security, defensive automation, and research.

This repository is for security practitioners, AI engineers, researchers, educators, and authorized red teams. It is deliberately broad: the aim is to explain what a resource does, why it matters, what evidence supports it, and where it belongs in a responsible workflow.

Responsible Use

All practical security work must be explicitly authorized, legally compliant, and proportionate to the agreed scope. This repository supports research, education, hardening, validation, detection, remediation, disclosure, and authorized testing. It does not provide instructions for unauthorized access, evading protections, deploying malware, operating command-and-control, impersonating people, or targeting third-party systems.

If you cannot demonstrate authorization for a target, do not test it. Use a lab, a benchmark, a capture-the-flag environment, or an intentionally vulnerable training application instead.

For coordinated disclosure guidance, consult the OWASP Vulnerability Disclosure Cheat Sheet and CERT/CC.

Contents

How to Navigate

The collection follows two complementary paths. AI for security concerns how AI can improve code review, vulnerability management, detection engineering, threat intelligence, incident response, and authorized assessments. Security for AI systems concerns models, data, RAG, agents, memory, tools, MCP servers, and their software supply chain.

Reader goalStart withThen continue with
Secure an LLM, RAG, or agentic applicationLLM, RAG, and Agent SecurityMCP Security and AI Security Guide
Evaluate security agents responsiblyAuthorized Security AgentsBenchmarks and Agentic Security Guide
Specialize a model for a security taskModels, Data, and AdaptationModels, Data, and Evaluation Guide
Build an approved MCP workflowMCP SecurityMCP server directory and MCP Security Guide
Learn systematicallyLearning PathFoundations, extended guides, benchmarks, and labs

Foundations, Governance, and Threat Taxonomies

Security work involving AI benefits from a shared vocabulary, a lifecycle view, and an evidence model. The following resources help teams reason about risk before selecting a tool or model.

ResourceTypeWhy it matters
OWASP GenAI LLM Top 10GuidanceCurrent OWASP guidance for LLM and generative AI application risks; the former site is maintained as a historical entry point.1
OWASP AI Security and Privacy GuideGuidanceBroad security and privacy reference for AI-system design, deployment, and operation.
NIST AI RMFFrameworkVoluntary framework organized around Govern, Map, Measure, and Manage.2
NCSC Secure AI System DevelopmentLifecycle guidanceSecure design, development, deployment, and operation/maintenance for AI systems.3
OWASP AI Security Verification Standard (AISVS)Verification standardVersioned, testable security requirements across data, models, infrastructure, identities, agents, MCP, adversarial robustness, and monitoring.20
MITRE ATLASThreat knowledge baseLiving knowledge base of threats to AI-enabled systems, with techniques, mitigations, and case studies.4
NIST Adversarial ML TaxonomyTaxonomyShared terminology for adversarial ML threats and mitigations.
AI Incident DatabaseIncident resourceDocumented AI incidents to inform risk discovery and governance exercises.
AVID TaxonomyRisk taxonomyStructured vocabulary for AI risk identification and documentation.
MLSecOps Top 10Practice guideSecurity perspective on the ML lifecycle and operational controls.

Security for AI Systems

LLM, RAG, and Agent Security

An AI application is more than a model. Prompts, retrieval sources, vector stores, identities, tool definitions, code execution paths, logs, memory, deployment infrastructure, providers, and users are all part of the security boundary. Secure design should consider the whole system.2 3

ResourceCategoryDescription
OWASP GenAI Security ProjectCommunity guidanceActive home for LLM, agentic AI, red teaming, incident response, and data-security resources.
PromptfooTesting frameworkDeclarative evaluation, red teaming, and regression testing for LLM applications, agents, and RAG systems.
GarakLLM probingOpen-source LLM vulnerability scanner for systematic model-behavior probing.
PyRITAI risk assessmentMicrosoft framework for identifying and mitigating generative-AI risks.
Purple LlamaSafety resourcesMeta tools and research artifacts for trust, safety, and LLM security evaluation.
FuzzyAIFuzzingAutomated LLM fuzzing for resilience and jailbreak testing.
Open Prompt InjectionBenchmarkBenchmark resources for studying prompt injection and defenses.
Agentic RadarAgent securityCLI scanner for issues in agentic workflows.
NVIDIA SkillSpectorSkill and extension securityScanner for agent skills and related supply-chain, permission, prompt-injection, and data-flow risks. Treat optional model-assisted analysis as an explicit data-handling decision.21
Agent Threat RulesDetection-rule formatMachine-readable agent-threat detection rules and reference implementations. The upstream format is a working draft; assess maturity and test coverage before enforcement.22
OWASP Agent Memory GuardMemory securityOWASP project for detecting and preventing agent memory poisoning.
Semgrep AI Best PracticesSecure developmentSecurity rules for code integrating LLM providers, MCP, assistants, and agent frameworks.
GiskardEvaluationOpen-source evaluation and testing for AI and LLM systems.
DeepTeamRed teamingLLM red-team testing for engineering workflows.

MCP Security

The Model Context Protocol (MCP) standardizes how hosts, clients, and servers expose resources, prompts, and tools to AI systems. That composability is powerful, but tools are security boundaries: they can access data or cause actions, and untrusted tool descriptions must be handled with caution.5

ResourceTypeDescription
MCP SpecificationProtocolCanonical concepts and trust-and-safety principles for resources, prompts, tools, authorization, and user interaction. Pin review to the deployed specification version.5
MCP Security Best PracticesTechnical guidanceCovers authorization, consent, token passthrough, SSRF, state handling, and local-server compromise.6
MCP Security Bench (MSB)Research benchmarkEnd-to-end evaluation of MCP-specific attacks across planning, invocation, and response handling. Use only in controlled evaluation and report both security and task-performance conditions.23
OWASP MCP Top 10Threat modelOWASP project in beta covering token exposure, scope creep, tool poisoning, supply chain, execution, and telemetry.7
OWASP Third-Party MCP GuideConsumer guidanceSecure discovery, authentication, authorization, sandboxing, least privilege, and human oversight.8
MCP Security ChecklistChecklistCommunity review aid; validate its advice against protocol guidance and the local environment.
Awesome MCP SecurityCurated listComplementary directory of MCP-security resources, tools, research, and guides.
Agent Scan (formerly mcp-scan)Security scannerTooling for inspecting MCP server exposures before adoption.
Cisco MCP ScannerSecurity scannerMulti-engine scanner for MCP servers and tools.

Before connecting a server, establish source provenance, transport, identities, requested scopes, filesystem and network access, whether it can read/write/execute, logging behavior, dependency posture, and approval steps for consequential actions. See the MCP Security Guide.

AI Red Teaming and Evaluation

AI red teaming is disciplined testing of models and AI applications under defined rules, controlled data, and measurable outcomes. Its purpose is remediation—not bypassing safeguards in deployed third-party systems.

ResourceFocusDescription
Adversarial Robustness ToolboxClassical MLIBM library for adversarial ML attacks, defenses, and robustness evaluation.
CounterfitClassical MLAutomation layer for assessing security risks in ML systems.
CleverHansClassical MLLibrary for adversarial examples and defense benchmarking.
FoolboxClassical MLRobustness-evaluation toolbox across common ML frameworks.
TextAttackNLP robustnessFramework for adversarial NLP testing, augmentation, and training.
AI Red Teaming Playground LabsTrainingLearning materials and labs for AI red teaming.
RAMPARTAgentic AI testingPytest-native framework for safety and security testing of agentic applications.
AgentDojoAgent-security benchmarkDynamic environment for prompt-injection attacks and defenses in tool-using agents. Use in an isolated test environment; its API is under active development.24
AI-Infra-GuardPlatformAI-security platform with infrastructure, MCP, and jailbreak evaluation components.

AI for Security

Application Security and Vulnerability Research

The high-value role of AI in AppSec is improving discovery, triage, validation, reporting, remediation, and learning while keeping people accountable for security decisions. Strong workflows combine model reasoning with deterministic checks, reproducible evidence, source-code context, and independent validation.9 10

ResourceAreaDescription
DARPA AI Cyber ChallengeVulnerability researchPublic effort around cyber reasoning systems for discovering and patching vulnerabilities in critical open-source software.
AIxCC Open Source ArchiveOpen systemsCompetition systems, artifacts, and resources released for study and defensive research.
AutoPatchBenchSecure repairBenchmarking for automated repair of fuzzing-detected vulnerabilities.
VLoc BenchLocalizationBenchmark for repository navigation, vulnerability localization, and patch verification.
VulnhallaCodeQL triageLLM-assisted triage for vulnerability-hunting workflows.
VulnHuntrCode reviewOpen-source vulnerability-hunting assistance for codebases.
SemgrepStatic analysisDeterministic companion to AI-assisted code review.
CodeQLCode analysisSemantic code analysis and query ecosystem.
OSS-FuzzFuzzingContinuous fuzzing for critical open-source projects.
SARIFInteroperabilityStandard format for static-analysis results and review.

The AIxCC final reported 54 synthetic vulnerabilities found out of 63 and 43 patched, alongside responsibly disclosed non-synthetic findings. The useful lesson is not a headline number: it is the importance of end-to-end validation, patch quality, and transparent scoring.11

Authorized Security Agents

There is no universal “best” security agent. Capability depends on task, target access, permitted tools, scaffold, model, evaluation protocol, and human oversight. The directory therefore groups resources by role and avoids global performance claims without comparable evidence.

ResourceRoleCuration note
PentAGIAuthorized pentesting agentMulti-agent system for complex security testing with sandboxing, multiple providers, knowledge-graph integrations, and observability. Review current permissions and deployment boundaries.
PentestGPTHuman-guided testingGPT-empowered assistance for penetration-testing research and authorized workflows.
HackingBuddyGPTResearch agentLLM-assisted security agent and benchmark resources.
ShannonWeb and API securityAutonomous application-security project; assess any performance claim in its stated test setting.
StrixApplication securityAgents for dynamic code analysis and validation.
Zen-AI-PentestFrameworkMulti-agent security testing framework; its tool integrations are privileged operations.
CyberStrikeAIPlatformAI-native tool orchestration, roles, and lifecycle management.
OpenHackCode securityMulti-agent source-code scanner and validation system.
CAI — archivedHistorical artifactThe project reports that it is archived and will receive no further fixes or security patches; retain as research context, not an active dependency.
BoxPwnrLab agentAgent resources for controlled security challenges.
NYU CTF AgentsResearch agentsAgents associated with scalable CTF evaluation research.
Cyber Security LLM AgentsAutomation examplesLLM-agent examples for common security workflows.

Strong agent designs separate architecture mapping, hypothesis generation, deterministic scanning, targeted testing, independent validation, evidence capture, deduplication, reporting, and remediation. A single agent that discovers and grades its own finding is a weak design.9 10

Threat Intelligence and Security Operations

AI can reduce toil in intelligence gathering, alert triage, detection engineering, and incident response. These workflows should default to read-only access, source citation, data minimization, analyst review, and explicit escalation paths.

ResourceAreaDescription
MITRE ATT&CKThreat knowledgeGeneral adversary-behavior knowledge base that complements ATLAS.
OpenCTICTI platformOpen platform for structuring and sharing threat knowledge.
MISPThreat sharingOpen-source threat-intelligence sharing platform.
SigmaDetection engineeringOpen generic signature format for shared detections.
YARAPattern matchingRule-based pattern matching for research and classification.
WazuhSecurity operationsOpen-source XDR/SIEM platform.
Elastic Detection RulesDetection engineeringPublic detection rules and rule-development resources.
Splunk Boss of the SOC DatasetTrainingDataset and environment for SOC-investigation practice.
CTI-BenchBenchmarkLLM benchmark focused on cyber-threat-intelligence tasks.
SECUREBenchmarkCybersecurity scenario dataset for extraction, understanding, and reasoning.

Models, Data, and Adaptation

Model specialization is not synonymous with training a model from scratch. RAG adds context at inference time without changing weights. Instruction tuning adapts behavior to instruction-response examples. PEFT/LoRA updates a small parameter subset. Domain-adaptive continuous pretraining adds domain knowledge through further pretraining. Distillation transfers behavior from a larger model to a smaller one. The right approach depends on objectives, data, permissions, risk tolerance, and evaluation.12

ResourceTypeWhy it belongs here
PrimusPaper, datasets, modelsCybersecurity resources spanning pretraining, instruction tuning, and reasoning distillation, with ablations and benchmark evaluation.13
Trend Micro Primus CollectionModels and dataCanonical collection referenced by the Primus paper.
Foundation-SecSpecialized modelCybersecurity-focused foundation model family; inspect model card, terms, data, and evaluation.
Antares CollectionSpecialized modelsSecurity-oriented models for vulnerability-localization workflows.
SecureBERTSpecialized modelBERT-family model for cybersecurity text tasks.
Llama-PrimusSpecialized modelCybersecurity model family associated with the Primus work.
Less Data, More SecurityResearchEmpirical study of domain-adaptive continuous pretraining with a curated corpus and multiple benchmarks.12
HackMentorHistorical projectRepository separating data construction, training, and evaluation for a cybersecurity LLM.
CyberLLMInstruct safety studySafety researchWhy cybersecurity-oriented fine-tuning needs safety evaluation before and after adaptation.14
CyberLLMInstruct (ACM AISec 2025)Dataset and paperPseudo-malicious cybersecurity instruction data and an empirical safety-performance trade-off study. Treat it as research material requiring local rights, risk, and pre/post safety review—not as a default training corpus.25
Model CardsDocumentationDocument intended use, provenance, limitations, evaluation, and ethical considerations.
Dataset CardsDocumentationDocument source, composition, license, processing, and limitations.

Any model, dataset, or adapter entry should state its license, source, intended use, limitations, data provenance, safety evaluation, benchmark protocol, and review date. The safety study found reduced safety resilience after fine-tuning in its evaluated settings; utility gains must not substitute for safety testing.14

Benchmarks, Cyber Ranges, and Evaluation

A benchmark score is meaningful only when task, model, agent scaffold, tools, environment, trials, feedback channels, budget, and metric are known. Do not compare white-box and black-box results, different numbers of attempts, hidden grader feedback, or contaminated tasks as though they were equivalent.

ResourceFocusDescription
CyBenchCyber capabilityProfessional CTF tasks with subtask measurement, metric definitions, logs, and cautions about comparability.15
CyberGymReal-world vulnerabilitiesLarge-scale benchmark across real vulnerabilities and projects; its abstract reports roughly 20% success even for the best evaluated combinations.16
NYU CTF BenchAgent evaluationDockerized CTF challenges for controlled LLM-agent evaluation.
CyberSecEvalModel evaluationCybersecurity capability and safety evaluation resources.
SecBenchSecurity knowledgeMulti-dimensional cybersecurity benchmark dataset.
CVE-BenchWeb securityVulnerable web applications and CVEs for controlled agent evaluation.
BountyBenchResearch impactBenchmark for vulnerability detection, validation, and patching.
AgentCyberRangeCyber-range researchOpen research infrastructure for evaluating agents in isolated multi-host web-exploitation and post-exploitation scenarios. Interpret reported scores only with stated prompts, budgets, and verification methods.26
AISI multi-step cyber-range studyCapability researchControlled study of multi-step cyber-range performance. Its findings show why budgets, repeated runs, containment, and task design need explicit reporting.27
AI Cyber Model ArenaAgent/model matrixWiz Research evaluation across real-world domains in isolated containers; interpret it as a specific methodology.17
BotsBenchSOC investigationEvaluation of agents on security-operations investigation tasks.
DefenseBenchDefensive agentsEvaluation of agents on defensive cybersecurity operations.

MCP Servers for Security Workflows

These are not plug-and-play endorsements. Prefer read-only access, minimal scopes, isolated environments, visible tool calls, and explicit human approval before any action that changes a system or sends traffic outside a lab.

Server or collectionAreaNotes
PortSwigger Burp Suite MCP ServerApplication securityOfficial integration between Burp Suite and MCP clients; constrain any configuration-changing capabilities.
MCP Security HubTool collectionDockerized collection; its documentation describes non-root containers, dropped capabilities, limits, health checks, and CI scanning. Inspect each server individually.18
mcp-for-securityCommunity collectionIndependently verify each upstream, package, and permission boundary.
OpenCTI MCP ServerThreat intelligenceBegin with a read-only service identity and audit access to intelligence data.
Wazuh MCP ServerSOC operationsSeparate telemetry queries from response or configuration capabilities.
Semgrep MCPCode securityIntegration for static-analysis workflows.
radare2 MCPBinary analysisUse isolated sample handling and read-only mounts.
VirusTotal MCPThreat intelligenceProtect API credentials and comply with data-sharing terms.
MCP ServersDiscoveryOfficial project collection and references; apply the same review to every server.

Research and Architecture Case Studies

Case studies are useful when read for design choices and stated limitations, not as independent rankings. The transferable lessons are task decomposition, independent validation, durable evidence, model diversity, controlled execution, and human decision ownership.

SourceFocusLesson
AWS Security AgentMulti-agent testingBaseline coverage, specialized workers, validation, structured evidence, and explicit recognition that plausible findings require rigorous validation.9
Cloudflare Vulnerability HarnessFleet-scale code securityExternal state, narrow agent context, separate hunting and validation, deduplication, and reproducible evidence before triage.10
Wiz AI Cyber Model ArenaEvaluation designSeparate model effects from agent effects, use ground truth, repeat trials, and isolate environments.17
Praetorian CVE ResearcherCVE automationResearch, technology correlation, detection design, critique, validation, and human review as distinct phases.19
Team AtlantaAIxCCFinalist retrospective and implementation-level learning.
Trail of Bits ButtercupAIxCCDefensive automated vulnerability-research and remediation case study.
Theori RoboDuckAIxCCFinalist retrospective and resources.
Agent security stack designPractitioner guidanceSeparate behavioral guidance from the runtime control plane: the agent proposes actions, while identity, policy, egress, and audit controls enforce allowed effects.28
AISI cyber-evaluation incident reportSafety case studyA controlled-evaluation incident that reinforces explicit egress decisions, real-time monitoring, solvable task scope, and containment that does not depend on model behavior.29

Learning Path

StageFocusSuggested starting resourcesOutcome
1. FoundationsApplication security, ML concepts, threat modelingOWASP, NIST, NCSC, ATT&CK/ATLASVocabulary for risks, assets, controls, and evidence.
2. Secure AI applicationsLLM, RAG, prompts, tools, and data flowsOWASP GenAI, Promptfoo, Garak, PyRITA test plan for an AI application and integrations.
3. Agents and MCPPermissions, memory, tools, isolation, logsMCP specification, MCP guidance, OWASP MCPA least-privilege architecture and review checklist.
4. AI-assisted AppSecCode analysis, fuzzing, validation, remediationSemgrep, CodeQL, OSS-Fuzz, AIxCCReproducible, reviewable findings.
5. AdaptationRAG, SFT, PEFT, domain adaptation, safetyPrimus, Foundation-Sec, DAP researchA documented retrieval-versus-tuning decision.
6. EvaluationBenchmarks, ranges, cost, failure analysisCyBench, CyberGym, VLoc BenchAn evaluation plan with limitations.
7. OperationsMonitoring, governance, disclosure, maintenanceNIST, NCSC, OWASPAn auditable program rather than a demo.

Extended Guides

GuideFocus
AI Security GuideSecuring LLM, RAG, and agentic applications across the lifecycle.
MCP Security GuideEvaluating and adopting MCP servers with secure permissions, identity, isolation, logs, and supply chain.
Models, Data, and Evaluation GuideResponsible cybersecurity-model specialization, dataset review, tuning, RAG, benchmarks, and release decisions.
Agentic Security GuideAuthorized AI-assisted security workflows with validation, evidence, autonomy boundaries, and human review.
Benchmarking and Evidence GuideEvaluation cards, comparable results, model-adaptation release gates, MCP-loop testing, cyber-range containment, and evidence reporting.
Curation PolicyInclusion rules, evidence standards, review cadence, deprecation, and contributor expectations.
Resource CatalogNormalized metadata, status, evidence, effects, and adoption boundaries for high-impact resources.
Resource Catalog SupplementMetadata and authorized-use boundaries for the remaining discovery resources promoted in this repository.
Catalog Maintenance GuideReproducible catalog regeneration, integrity checks, review authority, evidence requirements, and link-triage procedure.
Maintenance RegisterVisible review cadence, evidence to record, current baseline, and retirement rules.
Pre-Adoption ReviewRequired local verification before operational use of an unverified, experimental, historical, or high-impact resource.

Contributing and Curation

Contributions are welcome. A high-quality contribution gives readers enough information to assess relevance, activity, evidence, and safe use. Read CONTRIBUTING.md and the Curation Policy before opening an issue or pull request.

Every proposed entry should include a canonical URL, resource type, neutral description, maintainer, license, current status, relevant task or security domain, verification date, and evidence for quantitative claims. Entries involving tools, automation, or MCP must explain permissions, data exposure, actions, isolation, and authorized use. Archived, unmaintained, or experimental projects can be historically valuable, but must be labeled prominently.

References