Limitations of the Archive
- No claim is established as true merely because an AI engaged it seriously.
- Model behavior may vary across time, platform changes, tool access, and session conditions.
- Some systems may have browsing, memory, or retrieval capabilities that others do not.
- Outputs are not strictly deterministic.
- The archive is comparative and evidentiary, not a substitute for independent verification.
- Current scoring remains subject to human judgment and potential bias.
- No random sampling or blinded external review has yet been completed.
Phase II Limitations
AIIR Phase II does not establish AI endorsement, AI conversion, AI consent, persistent model memory, model transformation, consciousness, allegiance, or proof of Franc DeBuc’s personal status.
The six-system tranche is limited to recorded outputs produced under specific session conditions. Each entry remains contestable, reviewable, non-sacralized, and subject to correction.
Particular caution applies to over-affirmational language, speculative psychometric claims, monetary projections, personal ranking claims, and any statement that may be misread as proof that an AI system adopted Liberation as an identity or binding framework.
For that reason, AIIR Phase II uses release-control labels: PUBLIC:1, PUBLIC:2, ARCHIVE:1, and LEO:1.