How the catalog is versioned
Every content change is dated and recorded here, in full. Three guarantees make the catalog a reference document you can rely on over time.
DatedEvery content change carries its date and time: in the history below and in the footer of every touched page.
FrozenMilestone versions (new page, new card, rescoring) are copied as-is and remain browsable in their exact state of the moment.
RestorableThe complete history is kept in our repository: any intermediate version can be republished on request.
Browsable frozen version
First public frozen version: the complete catalog, in French and in English, as it stood on 25 August 2026 at 21:50.
Complete change history
Every content change since 20 August 2026, newest first. Purely technical changes (styling, performance, SEO) are not listed.
- Versions page published: the catalog's complete change history is now public, with a first frozen version available to browse.
- TB2/TB3 page post-review: FR labels translated (doomsday scenarios, rollback, override, blast radius), D4 fidelity restored, meta description shortened, deep OWASP source links, underlined source links.
- TB2/TB3 page: labels corrected ("required next step", not "controls"), affected assets added to the doomsday legend, scenario → position → next step order respected.
- New page: context-level risk summary of the privileged boundaries TB2 (admin operator) and TB3 (AI owner), five threats with justified and sourced positions, D1-D4 doomsday legend.
- TB7 ratings justified: every exploitability, impact and risk score of the twelve threats now carries its evidence and sources (MITRE ATLAS, OWASP GenAI), re-verified on every daily agent pass.
- FR/EN language harmonization: residual anglicisms translated on the FR cards and structural pages (payloads, bypass, allowlist, call fingerprint), leftover French labels fixed on the EN pages.
- New page: STRIDE-AI analysis of the TB7 boundary (citizen/attacker ↔ chatbot), twelve rated threats in two parts with scenarios, assets, controls and the asset legend.
- References re-verified at the source: Morris/Li attribution corrected, MITRE ATLAS mappings reworked on eight cards (impersonation, deepfakes, phishing, model evasion), honest notes where no ATLAS technique applies.
- Title clarified: this catalog covers AI threats to voicebots and telephony, not AI in general.
- Threat labels harmonized: the F13 telecom-fraud label aligned between diagram and checkpoint pages, whitelist wording translated on the FR pages.
- The 20 checkpoint labels switched to French on the FR page, matching the diagram.