Beyond Training: A Feasibility Taxonomy for Inference-Time AI Governance

arXiv:2609.10105v1 Announce Type: new Abstract: Compute governance today is a governance of training: the thresholds, reporting requirements, and frontier-AI regimes now in force attach to training compute and treat the trained model as the regulatory unit. That picture is incomplete: capability increasingly migrates to the deployment stage through inference-time scaling, agentic scaffolding, and compression onto consumer hardware. This paper asks which mechanisms are available once the regulatory object shifts from the training run to the inference call. We develop a feasibility taxonomy of twenty inference-time mechanisms across monitoring, verification, and enforcement, each rated on a four-point readiness scale against a documented four-vendor evidence base. We then stress the taxonomy against a two-dimensional adversary model (three capability tiers crossed with four adversary roles) and map each mechanism to four governance scenarios (domestic regulation, bilateral or multilateral coordination, industry self-regulation, and compute-marketplace governance). Fifteen of the twenty mechanisms have commercial technical substrates in production today, although governance-grade assurance and adversarial robustness vary substantially. The adversary analysis shows that this readiness holds only against a cooperative deployer and a low-to-medium-capability user: no mechanism rates adequate against a high-capability state-level deployer, and fine-tuning removes the model-internal components of the enforcement cluster, although platform-external controls can persist. A substituti

The paper presents a taxonomy of inference-time AI governance mechanisms and evaluates their readiness, but the analysis is based on a documented four-vendor evidence base and a two-dimensional adversary model, which may not be comprehensive. The results show that while some mechanisms are available, their effectiveness varies and none are adequate against a high-capability state-level deployer. The paper's focus on the feasibility of governance mechanisms rather than their actual impact on society makes the implications for human flourishing uncertain.

The paper highlights limitations of current AI governance mechanisms, particularly against high-capability adversaries.

The paper reveals a significant governance gap: current inference-time controls are insufficient against powerful state actors, raising risk of harmful AI use. While it supplies a useful taxonomy for future regulation, the immediate implication is a negative shift in human flourishing due to potential misuse.

SOURCE     arXiv cs.CY (Computers and Society) (tier 1)
   ↓
DOCUMENT   e3a997ec-e522-447c-990d-2d90169f45be
           https://arxiv.org/abs/2609.10105
   ↓
EVIDENCE   5 extracted excerpts
   ↓
MODEL RUN  3 runs, methodology 1.0.1
   ↓
SCORE      -29  (Adverse)
   ↓
CONFIDENCE 79%