~ / methodology
Methodology · v1.7.0

How we curate.

Yuntona is built on practitioner-led continuous intelligence gathering with expert curation — not automated scraping or vendor self-submission.

last reviewed · 2026-03-18 · single maintainer · accountable · MIT · open source
01

Philosophy

The AI security tooling landscape moves faster than any automated system can meaningfully evaluate. New tools, frameworks, and attack surfaces emerge weekly. Keeping pace requires continuous immersion in the practitioner community — not periodic desk research.

Every entry has been individually evaluated by a security professional with over a decade of experience spanning SOC/NOC operations, third-party risk, financial services regulation, and hands-on AI-security research. The directory reflects professional judgement informed by real-world operational context — not algorithmic scoring.

Curation quality depends on accountability, not volume.
02

Curation principles

Four commitments that shape every decision, from discovery to inclusion to retirement.

P.01
Multi-source intelligence
Discovery draws from practitioner discourse on LinkedIn and specialist forums, conference proceedings, standards body publications, OWASP working groups, industry intelligence newsletters, and direct engagement with AI-security thought leaders.
P.02
Expert signal filtering
Not everything discovered gets listed. Each candidate is assessed against the current threat landscape, mapped to OWASP categories, and evaluated for genuine operational utility. Duplicates without meaningful differentiation are excluded.
P.03
Community-embedded discovery
The best tools surface in comment threads, conference hallway conversations, and working-group Slacks — not product launch pages. Following the people who build and break AI systems is how tools reach us before mainstream awareness.
P.04
Continuous, not periodic
The directory updates on a rolling basis. This is not a quarterly report. It is a living artefact maintained through daily engagement with the AI-security ecosystem.
03

Intelligence sources

Where the signal comes from. Five channels feeding one pipeline.

Standards bodies
OWASPNISTCSAMITREISO/IEC
Publications + working group output
Industry intelligence
CB Insightsanalyst reportsmarket research
Vendor-neutral coverage of emerging tech
Conferences & events
NHIconCSA summitsAI security workshops
Vendor-neutral industry events
Practitioner networks
OWASP AIUCLinkedIn groupsspecialist forums
Thought-leader commentary
Primary research
academic papersarXivvendor docsGitHub
First-hand documentation
Pipeline
sources
5 channels
discovery
raw signal
evaluation
3 criteria
mapping
OWASP fit
published
directory
04

OWASP risk mapping

Every tool is mapped against the OWASP LLM Top 10 (2025) and the OWASP Agentic Top 10 (2026). Mappings are derived from each tool's documented capabilities, target threat model, and operational scope — assessed against the published risk descriptions.

LLM Top 10 · 2025
10 / 10 Full coverage
Agentic Top 10 · 2026
10 / 10 Full coverage
AI-assisted, human-designed
OWASP risk mappings across 161 tools and 20 categories were produced using AI as an analytical engine. The methodology, evaluation schema, and framework inputs were designed by the curator. Every output was reviewed and validated against the published OWASP standards. This is human-directed analysis at scale — not automated classification.
05

What gets listed

A tool is added when it meets three criteria. All three. No two-out-of-three.

C.1
Addresses a genuine security risk
in the generative or agentic AI stack, mapped to at least one OWASP category
C.2
Operational or near-operational
not vaporware, not a concept paper, not a GitHub readme without code
C.3
Offers meaningful capability
not already covered by existing entries; differentiated in scope, approach, or depth

Open-source tools, commercial platforms, and frameworks are all eligible. Vendor sponsorship does not influence inclusion or risk ratings.

06

Transparency

Yuntona is maintained by a single practitioner. The directory reflects one expert's informed judgement — not committee consensus, not crowd-sourced voting. This is a deliberate choice.

Affiliation disclosure

Yuntona is an independent, open-source project. It is not affiliated with or endorsed by OWASP, NIST, MITRE, or any vendor listed in the directory.

OWASP risk categories are used under the Creative Commons licence.