Hidden APIs in Language Models: Discovering Reusable Causal Interfaces from Forked Futures

2026-07-29

AI Summary

This work addresses the limitation of existing probing methods, which struggle to identify reusable internal causal interfaces in language models that support diverse future computations due to their reliance solely on current outputs. The authors propose a label-free framework for discovering such causal interfaces by introducing a “forked futures” mechanism: after a shared prefix, multiple divergent future trajectories are sampled to construct a causal quotient space based on comparisons of response distributions. Four interface types—including Shared—are defined and competitively evaluated using a preordered causal description length, with structural selection guided by a fidelity constraint on future signatures. Experiments demonstrate that the method reduces description length by 0.216 and 0.294 nats on Qwen2.5-1.5B and Llama-3-8B, respectively; it successfully recovers 14 out of 16 model architectures in blind tests and achieves an API alignment path mediation effect of 0.749.

Abstract

Identical language-model answers can arise from hidden states that support different future computations, so current-answer probes do not establish a reusable internal interface. We introduce forked futures: future operations are sampled only after a prefix state has formed, and states are compared through the response distributions induced by those operations. This yields an empirical causal quotient over hidden states without requiring researcher-specified latent labels. Shared, Local, Mixture, and Distributed interfaces then compete under prequential causal description length subject to future-signature fidelity and matched capacity constraints. In the two detailed model evaluations, Shared has the lowest held-out description length, with gains of 0.216 nats on Qwen2.5-1.5B and 0.294 nats on Llama-3-8B, while maintaining tightly clustered mean future-signature distortion; a five-backbone sweep preserves the positive direction of Sharedness Gain. The figure-aligned transplantation analysis gives Shared the strongest joint target-correctness, locality, copy-preservation, and composite profile, and API-aligned paths mediate 0.749 of the target effect versus 0.150 for matched null paths. In the blind four-class model-organism test, 14/16 architectures are recovered, with one observed non-Shared to Shared error among 12 non-Shared organisms. These results support an economical reusable causal interface within the tested operation banks, while keeping the claim explicitly conditional on the candidate architectures, interventions, and held-out futures.

Problem

• hidden APIs
• causal interfaces
• forked futures
• reusable interfaces
• language models

Innovation

• forked futures
• causal interface
• prequential description length
• future-signature fidelity
• hidden API