Sources and attribution

Inspected 2026-10-23. Public prose is paraphrased briefly; no payload corpus, slide deck, private report or upstream implementation is bundled.

S1 — Mapping

Jason Haddix, TBHM. Inspected Philosophy, Discovery, Mapping, Authorization/Session and testing checklist in the repository study. Checklist also credits The Web Application Hacker's Handbook. No license file found in that inspected tree; original synthesis only.

S2 — Evidence and coverage

JP Villanueva and Jason Haddix, HUNT. README covers passive parameter leads, manual methodology and scope. Apache-2.0 reported by repository/README. A lead is not a reproduced vulnerability; no tool executed.

S3 — AI-assisted investigation

Jason Haddix, Building AI Hackbots, Part 1, 2025-09-05. Read article sections on context, task scope, verification and human involvement. Its experience reports are author claims, not an independent comparison. We adopt concise evidence-linked rationale rather than its request for internal chain of thought. No general efficacy claim is imported.

S4 — Learning

Jason Haddix, Start Hacking LLMs, 2024-08-05. Read public article and controlled-learning suggestions. Subscriber interview not accessed. Historical recommendations do not establish current pricing or lab availability.

S5 — Surrounding infrastructure

Jason Haddix, The New Perimeter, 2026-05-01. Read AI development-stack inventory and defensive-posture discussion. We use the scoping idea, not its incident/CVE claims, which were not independently verified.

S6 — Prompt-injection vocabulary

Jason Haddix and Arcanum Information Security, PITAX, README declares v1.6.1. Inspected README and LICENSE.md, plus the author's release article, 2026-07-08. The README describes four pillars and a machine-readable taxonomy; individual JSON nodes and example payloads were not audited or imported.

Based on the Arcanum Prompt Injection Taxonomy by Jason Haddix, Arcanum Information Security (arcanum-sec.com), CC BY 4.0. Changes here: the four-axis vocabulary is adapted into a hypothesis worksheet; this is not the taxonomy or a claim of endorsement.

S7 — AI evaluation

Reversec Labs, spikee, README declares v0.9.2, Apache-2.0. Inspected docs 04, 05, 06, 09, 10, 11 and 15; canary judge implementation. The jhaddix repository is a fork, not evidence of his authorship. No spikee installation, provider call or real target test was performed for this package. See the wiki's evaluation study for version-specific limits.

S8 — Interview structure

Matt Pocock, grilling. Inspected decision-tree rounds and waiting for prerequisite answers. Our adaptation limits rounds to three useful questions, skips interviewing for definitions, and does not require delegation or grant execution permission. Not a Jason Haddix method.

Media and evidence limits

The HackTheBay 2025 organizer abstract and Hacker Valley episode page were inspected as a media index. Audio/video was not reviewed; no unseen talk's complete methodology is reconstructed from its title. The profile is an original working style based on the sources above, not a psychological model, biography or certification.