Fortress or Room? Metaphor, Method, and the Epistemology of Human–LLM Interaction Beyond Red Teaming

Fortress or Room? is a pillar document in the Crimson Hexagonal Archive’s operative-semiotics and Phase X sequence. It examines the metaphors used in LLM red teaming and asks what forms of knowledge become difficult to perceive when the model is understood primarily as fortress, captive, vehicle, material, landscape, or demon.

The paper begins by defending red teaming’s actual domain: adversarial testing discovers harmful outputs, policy leakage, and security weaknesses. Its argument is against universalization, not against the practice. When a specialized security method becomes the default ontology for every human–model encounter, collaborative and capacity-discovery phenomena may be interpreted only as failed safety, manipulation, or naïve attachment.

As an alternative method, the authors define logotic programming as structured interaction under explicit human-bearing constraints. The human supplies interpretation, commitment, and judgment; the model supplies constrained combinatorial synthesis. Roles, sources, operations, and outputs are recorded rather than left as informal prompt craft. The associated witness-teaming protocol is intended to make collaborative interaction reproducible and open to criticism.

The paper compares paradigms across ontology, purpose, technique, success conditions, power, and evidence. It also names risks in its own proposal, including projection of agency, self-sealing interpretation, unearned authority, and confusion between generated fluency and verified knowledge. Its conclusion is methodological: red teaming, alignment, instrumental prompting, and collaborative witness practices answer different questions and should not be collapsed into one adversarial grammar.