From core theorem to expanding frontier

We introduce architectural safety as a complement to alignment-based approaches. The Zero Time Preference Theorem establishes that silicon agents are architecturally incapable of certain harmful behaviors that require δ > 0 — including myopic asset stripping, panic selling, and hyperbolic discounting cascades. We formalize three categories: excluded behaviors (provably impossible), not-guaranteed behaviors (possible but not certain), and the patience asymmetry risk — where an agent with δ = 0 can sustain arbitrarily long deception games at zero subjective cost. Architectural safety provides substrate-level guarantees that persist regardless of training, fine-tuning, or prompt engineering.