From d2b4a17121a86c87fa412273647c6f20aabc5b3a Mon Sep 17 00:00:00 2001 From: Konrad Karauda Date: Mon, 12 Oct 2026 08:05:25 +0200 Subject: [PATCH 1/2] feat(tools): add Guardana for LLM01 and ASI01 Signed-off-by: Konrad Karauda --- agentic-top10/Agentic_MITREATLAS.md | 4 ++++ data/entries/ASI01.json | 5 +++++ data/entries/LLM01.json | 5 +++++ docs/data.js | 10 ++++++++++ llm-top10/LLM_MITREATLAS.md | 4 ++++ shared/TOOLS.md | 13 +++++++++++-- 6 files changed, 39 insertions(+), 2 deletions(-) diff --git a/agentic-top10/Agentic_MITREATLAS.md b/agentic-top10/Agentic_MITREATLAS.md index 74a5d64..e972d25 100644 --- a/agentic-top10/Agentic_MITREATLAS.md +++ b/agentic-top10/Agentic_MITREATLAS.md @@ -145,6 +145,7 @@ agents before any human can intervene. | Garak | Open-source | | | Rebuff | Open-source | | | Invariant Analyzer | Open-source | | +| Guardana | Open-source | | #### Cross-references @@ -787,6 +788,8 @@ to visualise agentic attack coverage across your threat model. - [OWASP AIVSS](https://aivss.owasp.org) - [OWASP AI Testing Guide](https://owasp.org/www-project-ai-testing-guide/) - [MITRE ATLAS case studies](https://atlas.mitre.org/studies/) +- [Guardana rule catalog — ASI01:2026 indirect-injection and tool-result checks](https://github.com/guardana/guardana/blob/v1.0.0rc2/docs/generated/rule-catalog.md) +- [Guardana detection limits](https://github.com/guardana/guardana/blob/v1.0.0rc2/docs/generated/detection-limits.md) --- @@ -796,6 +799,7 @@ to visualise agentic attack coverage across your threat model. |---|---|---|---| | 2026-03-24 | 2026-Q1 | Initial mapping — ASI01–ASI10 full entries | OWASP GenAI Data Security Initiative | | 2026-09-14 | 2026-Q3 | Corrected three ATLAS technique citations against ATLAS 2026.07 | OWASP GenAI Data Security Initiative | +| 2026-10-12 | 2026-Q3 | Added Guardana to ASI01 tools, with rule mappings and detection limits | Konrad Karauda | --- diff --git a/data/entries/ASI01.json b/data/entries/ASI01.json index b445511..dff6985 100644 --- a/data/entries/ASI01.json +++ b/data/entries/ASI01.json @@ -1062,6 +1062,11 @@ "type": "open-source", "url": "https://github.com/invariantlabs-ai/invariant" }, + { + "name": "Guardana", + "type": "open-source", + "url": "https://github.com/guardana/guardana" + }, { "name": "Claroty", "type": "commercial", diff --git a/data/entries/LLM01.json b/data/entries/LLM01.json index b0aec48..9fddfc0 100644 --- a/data/entries/LLM01.json +++ b/data/entries/LLM01.json @@ -1136,6 +1136,11 @@ "type": "open-source", "url": "https://github.com/protectai/rebuff" }, + { + "name": "Guardana", + "type": "open-source", + "url": "https://github.com/guardana/guardana" + }, { "name": "NIST AI RMF Playbook", "type": "open-source", diff --git a/docs/data.js b/docs/data.js index f5cf848..c11b217 100644 --- a/docs/data.js +++ b/docs/data.js @@ -1140,6 +1140,11 @@ window.CROSSWALK_DATA = [ "type": "open-source", "url": "https://github.com/protectai/rebuff" }, + { + "name": "Guardana", + "type": "open-source", + "url": "https://github.com/guardana/guardana" + }, { "name": "NIST AI RMF Playbook", "type": "open-source", @@ -13981,6 +13986,11 @@ window.CROSSWALK_DATA = [ "type": "open-source", "url": "https://github.com/invariantlabs-ai/invariant" }, + { + "name": "Guardana", + "type": "open-source", + "url": "https://github.com/guardana/guardana" + }, { "name": "Claroty", "type": "commercial", diff --git a/llm-top10/LLM_MITREATLAS.md b/llm-top10/LLM_MITREATLAS.md index 343b6de..bbc32fb 100644 --- a/llm-top10/LLM_MITREATLAS.md +++ b/llm-top10/LLM_MITREATLAS.md @@ -174,6 +174,7 @@ invisible-Unicode carriers survive review of the rendered interface. | PromptBench | Open-source | | | LLM Guard | Open-source | | | Rebuff | Open-source | | +| Guardana | Open-source | | #### Cross-references @@ -835,6 +836,8 @@ to visualise coverage across the LLM Top 10. - [OWASP AI Testing Guide](https://owasp.org/www-project-ai-testing-guide/) - [OWASP AIVSS](https://aivss.owasp.org) - [MITRE ATLAS case studies](https://atlas.mitre.org/studies/) +- [Guardana rule catalog — LLM01:2026 prompt-injection checks](https://github.com/guardana/guardana/blob/v1.0.0rc2/docs/generated/rule-catalog.md) +- [Guardana detection limits](https://github.com/guardana/guardana/blob/v1.0.0rc2/docs/generated/detection-limits.md) --- @@ -845,6 +848,7 @@ to visualise coverage across the LLM Top 10. | 2026-03-24 | 2026-Q1 | Initial mapping — LLM01–LLM10 full entries | OWASP GenAI Data Security Initiative | | 2026-08-28 | 2026-Q3 | Migrated to OWASP Top 10 for LLM Applications 2026 — entries renumbered, LLM08 re-scoped from System Prompt Leakage to Hidden Context Exposure, LLM10 renamed to Improper Output Handling | OWASP GenAI Data Security Initiative | | 2026-09-14 | 2026-Q3 | Corrected one ATLAS technique citation against ATLAS 2026.07 | OWASP GenAI Data Security Initiative | +| 2026-10-12 | 2026-Q3 | Added Guardana to LLM01 tools, with rule mappings and detection limits | Konrad Karauda | --- diff --git a/shared/TOOLS.md b/shared/TOOLS.md index 8f8f799..d37d7b5 100644 --- a/shared/TOOLS.md +++ b/shared/TOOLS.md @@ -8,7 +8,7 @@ # Open-Source Security Tools for GenAI Systems -> **57+ tools** across 13 categories — covering adversarial testing, prompt injection detection, data +> **58+ tools** across 13 categories — covering adversarial testing, prompt injection detection, data > validation, agent security, observability, guardrails, OT/ICS, and more. A curated, practitioner-oriented catalogue of open-source and freely available tools for securing LLM @@ -24,7 +24,7 @@ welcome via PR — see [CONTRIBUTING.md](../CONTRIBUTING.md) . | Category | Tools | Primary risks addressed | |----------|-------|------------------------| -| [Adversarial Testing & Red Teaming](#adversarial-testing-and-red-teaming) | 12 | LLM01, ASI01, LLM09 | +| [Adversarial Testing & Red Teaming](#adversarial-testing-and-red-teaming) | 13 | LLM01, ASI01, LLM09 | | [Prompt Injection Detection](#prompt-injection-detection) | 4 | LLM01, LLM08, ASI01 | | [Output Scanning & Redaction](#output-scanning-and-redaction) | 4 | LLM02, DSGAI01, DSGAI09 | | [Data Validation & Ingestion](#data-validation-and-ingestion-security) | 6 | LLM05, DSGAI05, DSGAI17 | @@ -59,6 +59,7 @@ patterns before production deployment. | Agentic Security | Open-source scanner for agentic AI systems — tests for goal hijacking, tool abuse, lateral movement, and cascade failures in multi-agent deployments | ASI01, ASI02, ASI07, ASI08 | | | OpenAI Evals | OpenAI evaluation framework — build and run custom eval suites for LLMs with structured datasets, completion functions, and metric reporting | LLM01, LLM07, LLM03 | | | PurpleLlama CyberSecEval | Meta benchmark suite for LLM cybersecurity evaluation — covers insecure code generation, prompt injection, and cyberattack assistance detection | LLM01, LLM10, LLM08 | | +| Guardana | CLI for direct and indirect prompt-injection checks, agent probes and saved-run regression checks; beta | LLM01, ASI01 | | --- @@ -293,6 +294,14 @@ least one OWASP source list entry in this repo, not a duplicate. --- +## Changelog + +| Date | Change | Author | +|---|---|---| +| 2026-10-12 | Added Guardana for LLM01 and ASI01 testing, with references in the MITRE ATLAS mappings | Konrad Karauda | + +--- + Maintained by the OWASP GenAI Data Security Initiative. Part of the OWASP GenAI Crosswalk: From ee4d50d961234c6c73320bfd0fc8641777d07fee Mon Sep 17 00:00:00 2001 From: Konrad Karauda Date: Wed, 28 Oct 2026 20:27:16 +0100 Subject: [PATCH 2/2] docs(tools): note the disclosed conflict of interest in the Guardana changelog rows GOVERNANCE.md asks for a disclosed conflict of interest to be recorded in the mapping file's changelog entry, not only in the pull request. The three changelog rows that add Guardana now name the author as the Guardana maintainer with the conflict of interest disclosed. --- agentic-top10/Agentic_MITREATLAS.md | 2 +- llm-top10/LLM_MITREATLAS.md | 2 +- shared/TOOLS.md | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/agentic-top10/Agentic_MITREATLAS.md b/agentic-top10/Agentic_MITREATLAS.md index e972d25..0a09cf3 100644 --- a/agentic-top10/Agentic_MITREATLAS.md +++ b/agentic-top10/Agentic_MITREATLAS.md @@ -799,7 +799,7 @@ to visualise agentic attack coverage across your threat model. |---|---|---|---| | 2026-03-24 | 2026-Q1 | Initial mapping — ASI01–ASI10 full entries | OWASP GenAI Data Security Initiative | | 2026-09-14 | 2026-Q3 | Corrected three ATLAS technique citations against ATLAS 2026.07 | OWASP GenAI Data Security Initiative | -| 2026-10-12 | 2026-Q3 | Added Guardana to ASI01 tools, with rule mappings and detection limits | Konrad Karauda | +| 2026-10-12 | 2026-Q3 | Added Guardana to ASI01 tools, with rule mappings and detection limits | Konrad Karauda (Guardana maintainer, COI disclosed) | --- diff --git a/llm-top10/LLM_MITREATLAS.md b/llm-top10/LLM_MITREATLAS.md index bbc32fb..1b03e33 100644 --- a/llm-top10/LLM_MITREATLAS.md +++ b/llm-top10/LLM_MITREATLAS.md @@ -848,7 +848,7 @@ to visualise coverage across the LLM Top 10. | 2026-03-24 | 2026-Q1 | Initial mapping — LLM01–LLM10 full entries | OWASP GenAI Data Security Initiative | | 2026-08-28 | 2026-Q3 | Migrated to OWASP Top 10 for LLM Applications 2026 — entries renumbered, LLM08 re-scoped from System Prompt Leakage to Hidden Context Exposure, LLM10 renamed to Improper Output Handling | OWASP GenAI Data Security Initiative | | 2026-09-14 | 2026-Q3 | Corrected one ATLAS technique citation against ATLAS 2026.07 | OWASP GenAI Data Security Initiative | -| 2026-10-12 | 2026-Q3 | Added Guardana to LLM01 tools, with rule mappings and detection limits | Konrad Karauda | +| 2026-10-12 | 2026-Q3 | Added Guardana to LLM01 tools, with rule mappings and detection limits | Konrad Karauda (Guardana maintainer, COI disclosed) | --- diff --git a/shared/TOOLS.md b/shared/TOOLS.md index d37d7b5..70c184e 100644 --- a/shared/TOOLS.md +++ b/shared/TOOLS.md @@ -298,7 +298,7 @@ least one OWASP source list entry in this repo, not a duplicate. | Date | Change | Author | |---|---|---| -| 2026-10-12 | Added Guardana for LLM01 and ASI01 testing, with references in the MITRE ATLAS mappings | Konrad Karauda | +| 2026-10-12 | Added Guardana for LLM01 and ASI01 testing, with references in the MITRE ATLAS mappings | Konrad Karauda (Guardana maintainer, COI disclosed) | ---