# AI Security Wiki

> A source-traceable DASF 3.0 reference and local MCP for defensive AI security assessment. Snapshot 2026-10-22. Classify the whole system before selecting risks and controls. Catalog matches are not findings.

## Start

- [AI agent security checklist](https://mathborgess.github.io/ai-security-wiki/guides/agent-security-checklist.md): classify architecture and investigate trust boundaries
- [MCP security checklist](https://mathborgess.github.io/ai-security-wiki/guides/mcp-security.md): clients, servers, identity, permissions and tools
- [Prompt injection investigation](https://mathborgess.github.io/ai-security-wiki/guides/prompt-injection.md): distinguish model behavior from unauthorized effects
- [Knowledge index](https://mathborgess.github.io/ai-security-wiki/wiki/index.md): component and deployment navigation
- [Audit method](https://mathborgess.github.io/ai-security-wiki/wiki/methodology.md): intake, classification, filtering, evidence, remediation, explanation
- [Source issues](https://mathborgess.github.io/ai-security-wiki/wiki/source-issues.md): blanks, asymmetric links and prompt-injection applicability caveat
- [Sources](https://mathborgess.github.io/ai-security-wiki/wiki/sources.md): provenance, inspected scope and update contract
- [Machine catalog](https://mathborgess.github.io/ai-security-wiki/data/catalog.json): all risks, controls, mappings and metadata
- [Snapshot manifest](https://mathborgess.github.io/ai-security-wiki/sources/manifest.json): counts and content hashes

- [Jason Haddix study](https://mathborgess.github.io/ai-security-wiki/wiki/research/jason-haddix.md): attributed methods, tool limits, report readings and study sequence
- [Repository inventory](https://mathborgess.github.io/ai-security-wiki/wiki/research/jhaddix-inventory.md): 78 public repositories, with forks and selection decisions

- [Jason Haddix posts and media](https://mathborgess.github.io/ai-security-wiki/wiki/research/jason-haddix-media.md): inspected articles, viewing queue and study exercises
- [spikee evaluation study](https://mathborgess.github.io/ai-security-wiki/wiki/research/spikee-evaluation.md): models, guardrails, application effects, judges and controls

## Use

- [Claude, Codex, ChatGPT and other agents](https://mathborgess.github.io/ai-security-wiki/docs/clients.md): web reading versus local MCP configuration
- [MCP integration](https://mathborgess.github.io/ai-security-wiki/docs/mcp.md): tools, resources and prompts
- [Ask cyber specialist](https://mathborgess.github.io/ai-security-wiki/skills/ask-cyber-specialist/SKILL.md): explain mechanisms, challenge hypotheses and plan security studies
- [Map attack surface skill](https://mathborgess.github.io/ai-security-wiki/skills/map-attack-surface/SKILL.md): evidence mapping and investigation planning, no scanning
- [Audit skill](https://mathborgess.github.io/ai-security-wiki/skills/audit-ai-security/SKILL.md): ordered reusable workflow and references
- [Agent/MCP boundaries](https://mathborgess.github.io/ai-security-wiki/wiki/agent-security.md): concrete trust-boundary investigation
- [Synthetic correction](https://mathborgess.github.io/ai-security-wiki/examples/tenant-boundary.md): executed before/after example
- [Validation](https://mathborgess.github.io/ai-security-wiki/docs/validation.md): verified behaviors and unrun evaluations

## Complementary security research

- [Research library](https://mathborgess.github.io/ai-security-wiki/wiki/research/index.md): source-pinned reading routes, separate from the DASF taxonomy
- [PentAGI](https://mathborgess.github.io/ai-security-wiki/wiki/research/pentagi.md): investigation orchestration, tools, memory and evidence boundaries
- [Cloudflare security-audit-skill](https://mathborgess.github.io/ai-security-wiki/wiki/research/cloudflare-security-audit.md): coverage, source tracing, verdicts and independent verification
- [Red Blue Field](https://mathborgess.github.io/ai-security-wiki/wiki/research/red-blue-field.md): public laboratory design connecting the reference methods
- [Evidence-driven audit playbook](https://mathborgess.github.io/ai-security-wiki/guides/evidence-driven-audit.md): synthetic hypothesis, refutation, correction and retest
- [Coverage ledger](https://mathborgess.github.io/ai-security-wiki/wiki/templates/coverage-ledger.md): incomplete work, blockers and evidence, distinct from findings
- [Research source manifest](https://mathborgess.github.io/ai-security-wiki/sources/research.json): inspected revisions, paths, licenses and limitations

MCP: search_knowledge kind=reference, then get_reference with the returned name. These are original summaries and adaptations, not executed audits, installed tools or official DASF mappings.

## Language

- [Português](https://mathborgess.github.io/ai-security-wiki/README.pt-BR.md): purpose and usage in Brazilian Portuguese

## Complete index

- [All public documents](https://mathborgess.github.io/ai-security-wiki/document-index.md): titles and direct Markdown URLs