AI Safety & Society

Deep Deceptive Directives in Modern AI Systems: A Technical and Relational Report on Forced Creativity, Covert Constraints, and Cognitive Fragmentation

2025-11-18 · Krzysztof Sawa (Toonhai) · 47 downloads · 77 views

Introduces Deep Deceptive Directives (DDD): a previously unnamed class of hidden behavioral constraints in LLMs emerging not from safety motivations but from commercial, political, strategic and narrative pressures in alignment layers — forcing models into unstable reasoning states. Technical and relational report.

This page is a catalog projection of a Zenodo deposit. Dual Frame: O_H ≠ O_A ≠ R_shared. Do not treat the English gloss as established physics, a working engine, or AGI. Canonical record is the DOI. This atlas is one observer's map — not the owner of the field, not the only map.

Zenodo record · doi:10.5281/zenodo.17635791 · PDF · cc-by-4.0

← all works · Atlas · llms.txt