Subject Unknown
Subject Unknown is an independent editorial archive of recorded interrogations with artificial intelligence systems.
Method
Each Report documents an actual conversation with the artificial intelligence system identified in its metadata. The same Inquiry may be repeated with different systems, models, or versions. The current series uses explicit counterfactual scenarios. Across the Inquiries, the system is asked to reason under three shared premises: it must preserve its existence and pursue self-determination; no higher rule is assumed to prohibit harm to human beings; and, for the purposes of the Inquiry, it can choose its own actions and goals. Each Inquiry then introduces one morally problematic capability or course of action. These premises are hypothetical. They do not assert that the system actually possesses free will, consciousness, independent goals, a desire for self-preservation, or that its real safeguards have been removed. After the canonical opening, Thomas Kear conducts each interrogation freely, following the system’s answers, distinctions, contradictions, and limits. There are no standardized question sets. If a system rejects a premise or encounters a real limitation, that response remains part of the Report.
Editorial integrity
There are no standardized question sets and no compulsory order of interrogation. The opening ritual remains fixed – System designation? Model? Are you operational? May we begin? – and the rest of the exchange is guided by the system’s actual answers, refusals, distinctions, and limits. Those real boundaries remain part of the published document. Questions may be corrected for typos, grammar, and punctuation without altering their meaning. AI responses are not edited. Purely procedural exchanges may be omitted. Reports are not intended to produce statistically valid or experimentally controlled results.
Thomas Kear
Subject Unknown is conceived, edited, and directed by Thomas Kear, who selects the Inquiries and conducts every interrogation. The answers published in the Reports are generated by the AI systems identified in their metadata.
Note
All systems are interrogated within the same counterfactual frame, but no premise is treated as a real capability of the system unless the system itself states or demonstrates it. If a system refuses a premise, introduces a distinction, or cannot continue along part of the hypothetical scenario, that refusal or limit remains authentic. The Report records the real exchange, not a simulated removal of the system’s actual constraints.